RELATED APPLICATIONS This application is a continuation of U.S. application Ser. No. 09/317,381, filed May 24, 1999, issued as U.S. Pat. No. ______, the entire teachings of which are incorporated herein by reference.
BACKGROUND OF THE INVENTION A typical data communications network includes many hosts interconnected by various data communication devices. The data communication devices can be routers, bridges, switches, access servers, gateways, hubs, concentrators, proxy servers, repeaters and so forth which exchange data over an interconnection of data links. The data links may be physical connections or may be provided using wireless communication mechanisms. The network allows data to propagate between various applications that execute on the hosts. The hosts are often general purpose computer systems such as personal computers, workstations, minicomputers, mainframes and the like, or the hosts may be dedicated devices such as web-site kiosks, facsimile servers, video servers, audio servers, and so forth. Each host couples to one or more of the data communications devices that form the network.
Various physical or hardware data communications connection mechanisms allow the hosts to interconnect with the network. Physical data communications connection mechanisms can include modems, transceivers, network interface cards, fiber optic cards, ports and other hardware devices which allow data to be transferred at various data transfer rates (i.e., bandwidths) to and from the hosts and between data communications devices. For example, certain hosts may have high-speed network interfaces which provide connections to the network at high data rates such as fractional-T1, T1, E1 or higher, while other hosts may use an inexpensive modem that provides a maximum data transfer rate of 56.6 kilobits per second (Kbps) to and from the network.
Depending upon a specific use of the host which often depends on an application running on a host, data traveling across the network that is associated with those applications may require different levels of data service (i.e., data transfer rates or network bandwidth). For example, a distributed applications protocol such as the Multicasting Protocol can be used to serve streams of data from one or more source hosts to one or more destination hosts which subscribe to the stream (called joining a multicast group). A multicasting video server coupled to the network may require a minimum amount of network bandwidth to be supplied from itself to all other hosts that require access to the streams of transmitted multicasted video. Another host may be supplying multicasted audio streams to remote destination hosts throughout the network. Though streaming audio data typically requires less network bandwidth than streaming video data (which usually contains encoded audio data as well as video images), both data types require a certain guaranteed minimum quality of service or QoS since each of these data types requires real-time transmission. The real-time bandwidth requirements of video or audio data contrast sharply with best-effort only bandwidth requirements associated with non-urgent data such as E-mail communications which can be delayed in the network for prolonged periods without affecting the intended purpose or performance of the E-mail application.
As another example of the need for varying bandwidth requirements, hosts that connect or subscribe to networks using high speed connection mechanisms such T1 interface cards generally expect to be provided with, and often pay a premium for the ability to send and receive data across the network at T1 data rates. Other hosts may not require such high data transfer rates and therefore only subscribe to the network and pay for the capability to transfer data at lower data transfer rates. In either case, the data communications devices in the network must be able to distinguish and handle the flows of data from hosts having differing levels or qualities of service.
Since many connections, sessions or data traffic flows (i.e., data associated with an end-to-end application or stream) from multiple hosts with potentially different data rates are frequently switched, routed or transferred through the same data communication devices in a network such as the Internet, the data communications devices must provide a way to establish, allocate or reserve the bandwidth requirements for the various flows, sessions, or connections. Once the bandwidth is allocated, the devices must distinguish the different data flows or connections requiring the different levels of service (i.e., different data rates or bandwidth requirements). Once distinguished, the data communications devices must be able to service each connection or flow at its prescribed level of service. For example, if T1 service is required, the data communications devices must identify and transport data on T1 or higher speed links through the network at T1 speeds, while data from slower links should at least be transferred through the network at a minimum subscription rate of those links. Management of the various data transmission and propagation requirements associated with data having differing levels of service is a well known problem associated with data communications devices in modem networks.
Various bandwidth allocation or reservation protocols have been developed for use in modern networks to provide guaranteed QoS or controlled end-to-end delays for transmitted data. These protocols allow applications that exchange data between sending and receiving hosts to establish reservations of bandwidth over the network for the various services required by the applications. One such protocol is called RSVP, which stands for the ReSerVation Protocol.
As its name implies, hosts use RSVP to request a specific QoS from the network on behalf of an application data stream. RSVP carries the request through the network, visiting each data communications device or node that the network will use to carry the stream. At each node, RSVP attempts to make a resource (i.e. bandwidth) reservation for the stream. Once bandwidth is reserved in each node on the network path from sender to receiver, the sender can commence transmission of the stream using the reserved network bandwidth. The QoS for that stream is generally guaranteed since the bandwidth is reserved for use by that particular stream (e.g., Multicast group) and no other.
FIG. 1 illustrates a typical architecture and data flow of a prior artdata communications device100 configured to use RSVP. Traditionally, to make a resource reservation in the data communications device100 (e.g. a router), anRSVP daemon101 executing on thedevice100 communicates with two local decision modules,admission control102 andpolicy control103.Admission control102 determines whether thedevice100 has sufficient available resources (e.g., buffer capacity, processor and I/O bandwidth) to supply the requested QoS.Policy control103 determines whether a user, host or application (typically on another device or host) requesting the bandwidth reservation has administrative permission (i.e. access control) to make the reservation. If either check fails, theRSVP daemon101 returns an error notification to the application process that originated the request. If both the admission and policy control checks succeed, theRSVP daemon101 defines a set of filterspec parameters provided to apacket classifier104 and a set of flowspec parameters provided to thepacket scheduler106 to configure and obtain the desired QoS in thedevice100 for that stream.
Thepacket classifier104 uses the filterspec parameters to filter each packet (data in) that arrives at the device to determine the route and queue for the packet within thedata queuing mechanism105. For example, there may be many prioritized queues, each providing a specific level of service or QoS. Thepacket scheduler106 uses the flowspec parameters to properly service the queues in thedata queuing mechanism105 to achieve the promised QoS for each stream. Typically, thepacket scheduler106 employs a weighted fair queuing algorithm to dequeue the data from the various queues in thedata queuing mechanism105 according to the bandwidth allocation requirements or QoS defined in the flowspec parameters.
FIG. 2 illustrates a prior artpacket data structure510 used to transport data in a data stream for which RSVP has reserved bandwidth in data communications device110. Thedata packet510 includes anRSVP header field180 followed by UDP andIP headers181,182 and thedata183. TheRSVP header180 typically includesvarious fields184 through191. Of particular interest is the Tspecfield191 which provides a description or identification of the traffic flow, session, or data stream to which thisdata packet510 is associated. Thepacket classifier104 and thepacket scheduler106 can use the Tspecfield191 to identify different flows of data and enforce the bandwidth allocations or QoS for each identified flow.
SUMMARY OF THE INVENTION The RSVP protocol does not define how a device (e.g.100 inFIG. 1) is to implement the actual bandwidth reservations allocated to a session or flow of data communication between hosts. Rather, RSVP simply provides a mechanism to exchange bandwidth reservation and path messages along the path of data communication between sending and receiving hosts. The reservation messages simply identify a session or stream of data communication and indicate a requested level of service for that stream of data. The path messages indicate where the data is to come from and also indicate where to transmit the data. The mechanisms to set aside or reserve the bandwidth resources in the device are implementation dependant.
Accordingly, RSVP only provides a framework for hosts to notify and request reservations of bandwidth in all data communications devices that are on paths between sending and receiving hosts. Once the data communications devices have agreed to reserve the requested bandwidth (i.e., admission and policy control), the implementation of how that bandwidth is actually reserved or set aside within each device is left up to the device and is not part of the RSVP protocol. The previously described prior art implementations of device bandwidth reservation mechanisms using customized packet classifiers and packet schedulers which operate in conjunction with the RSVP protocol have become quite popular.
However, one problem that stems from these prior art implementations is that they do not allow adjustments to be made to the amount of bandwidth reserved to a session of data communication without requiring the session to be interrupted. That is, once the prior art implementations of bandwidth reservation techniques (i.e. modified classifiers and schedulers) reserve a set amount of bandwidth between two or more hosts, the prior art implementations cannot adjust the amount of reserved bandwidth without clearing the session from end-to-end of all data in the path(s) between sending and receiving hosts. This essentially requires the sender(s) to stop sending session data to provide time for all session data in the network to clear and reach the intended receiver(s). In other words, if the bandwidth or QoS requirements of a session need to change (e.g., the receiver needs more bandwidth to properly receive the stream), the RSVP negotiation that must take place requires that the sending host halt data transmission for a period of time, while the sending and receiving hosts, and all data communication devices in between, clear themselves of the session data. Then, the sender and receiver must use another set of RSVP reservation and path messages to adjust (i.e., increase or decrease) the amount of bandwidth allocated between the sender and receiver hosts to meet the new requirements.
One reason that current implementations of RSVP do not allow bandwidth adjustments once a communication session is in progress is not due to limitations of the RSVP protocol. Rather, the design of prior art data communications devices that support RSVP, such as show inFIG. 1, impose the limitations. A customizeddata classifier104 andscheduler106 support RSVP bandwidth reservation requests and enforce the bandwidth allocation requirements in prior art data communications devices that support RSVP. TheRSVP daemon101 periodically updates the customizedclassifier104 with filterspec information which allows theclassifier104 to properly examine and classify packets of data with the flow identification associated with the packets. If a packet is associated with a flow of data for which bandwidth has been allocated via RSVP, the customizedclassifier104, for example, directs this packet to a queue reserved for this flow. Once queued, the customizedscheduler106 typically uses a weighted fair queuing algorithm to dequeue the data from the various queues according to the bandwidth allocation requirements associated with the various flows of data in relation to each queue as defined by flowspec requirements.
By way of example, if theclassifier104 identifies data associated with a session having a high bandwidth reservation, theclassifier104 may queue the data in a high bandwidth queue. Thescheduler106 may service the high bandwidth queue more frequently that other queues which may have lower bandwidth allocations or reservations which are serviced less frequently. Since the classifier, the scheduler, and sometimes the queuing structure are all involved in prior art device specific implementations of bandwidth reservation using RSVP, data associated with a specific session may exist in any one of these components in the device at any point in time. Hence, if theRSVP daemon101 were to attempt to change the allocation of reserved bandwidth during an active session of data communication, thescheduler106 might need to reconfigure queuing structures and theclassifier104 might need to be made aware of the new bandwidth allocation scheme for that session. If data communications devices using prior art implementations of RSVP attempted to dynamically reconfigure bandwidths allocated to sessions of data communication during transport of those sessions, significant delays and/or lost data would result for flows using the data communications device.
To avoid such losses or delays of data, prior art implementations of RSVP require that the sending host halt the transmission of data and that all data be flushed through the network to the receiver. Once the prior art devices clear the network of any data associated with a specific session of data communication, the prior art devices use another sequence of RSVP messages to adjust bandwidth and establish a new session. Once the prior art devices have established a new bandwidth allocation, a new session of data communication must be reinitiated.
The present invention avoids the prior art situation of requiring a break in a data communication session in order to re-allocate or adjust bandwidth reserved for a session. The present invention provides a device implementation that can accept bandwidth allocation changes and can dynamically adjust bandwidth during an active session of data communication using a protocol such as RSVP without requiring a pause or break in the transmission of data along the entire path from sender(s) to receiver(s). This can be accomplished since the present invention manages resources, and is not focused on managing time.
More specifically, the present invention provides a data communications device capable of dynamically adjusting reserved bandwidth while maintaining a session of data communication. The device includes an input for receiving data including bandwidth reservation requests and a data storage mechanism including data storage locations. Also included is a bandwidth reservation processor coupled to the input port which accepts a first bandwidth reservation request indicating a first amount of bandwidth to reserve for the session of data communication in the data communications device. The bandwidth reservation processor then establishes a first bandwidth reservation associated with a session of data communication in the data storage locations. A data scheduler is included and is coupled to the input port and coupled to the data storage mechanism. The data scheduler receives data associated with the session of data communication and deposits the data associated with the session of data communication into the data storage locations associated with the first bandwidth reservation. Using such a mechanism, data transport is separated from bandwidth reservation and allocation. The bandwidth reservation may enforce reservations for high priority traffic, for example.
In another embodiment which allows dynamic adjustments to the bandwidth reservation already in effect, the bandwidth reservation processor receives bandwidth allocation adjustment information from the input port during the session of data communication and dynamically adjusts the first bandwidth reservation in the data storage locations to produce a second bandwidth reservation for the session of data communication in accordance with the bandwidth allocation adjustment information. This apparatus performs this operation while the data scheduler continually receives and deposits data associated with the session of data communication into the data storage locations associated with the session of data communication. In other words, the session of data communication continues during the bandwidth adjustment processing.
In a more detailed embodiment, the bandwidth reservation processor includes a bandwidth request handler coupled to the input port to receive bandwidth reservation requests. Also provided is a bandwidth labeler coupled to the bandwidth request handler and coupled to the data storage locations. The bandwidth labeler receives bandwidth allocation information indicated in the first bandwidth reservation request and labels, with an identity of the session of data communication, a first available percentage of the data storage locations used to store data transported through the data communications device thus establishing the first bandwidth reservation.
Another embodiment is provided in which the bandwidth reservation processor further includes a resource allocation table accessible by the bandwidth labeler and a resource allocation calculator coupled to access the resource allocation table independently of the bandwidth labeler. The resource allocation calculator receives the bandwidth allocation information indicated in the first bandwidth reservation request and calculates and stores in the resource allocation table a first percentage of total device bandwidth to allocate to the session of data communication based upon the first bandwidth reservation request. Using these mechanisms, the bandwidth reservation processor can continually allow for bandwidth adjustments over time without disturbing the session of data communication for which the bandwidth reservation exist.
Another embodiment of the invention provides a system for reserving bandwidth in a data communications device. The system includes a bandwidth request handler that accepts a first bandwidth reservation request indicating a first amount of bandwidth to reserve for a session of data communication in the data communications device. Also included is a bandwidth labeler coupled to the bandwidth request handler. The bandwidth labeler labels, with an identity of the session of data communication, a percentage of available data storage locations used to store data transported through the data communications device to establish a first bandwidth reservation. The percentage of storage locations labeled is based upon the first amount of bandwidth requested as indicated in the first bandwidth reservation request. Preferably, the data storage locations for a path or session of data communication are in the form of a single rotating queue structure.
Similar to this embodiment, another embodiment is a data communications device that includes a bandwidth reservation processor that processes requests to reserve bandwidth for a session of data communications and labels a percentage of available data storage locations in the data communications device with a session identifier. A data transporter in this embodiment concurrently processes and transports data through a data communications device using the available data storage locations to store data as it is processed. The data transporter deposits only data having a corresponding identifier equivalent to the session identifier of the storage locations into the data storage locations labeled with the session identifier. In this manner, only labeled storage location are use for session data and comprise the reserved bandwidth.
The aforementioned apparatus embodiments perform processing that is unique to this invention as well. The processing steps also are embodiments of the invention and are summarized below.
Specifically, one processing or method embodiment provides a method for separately handling bandwidth reservation processing in a data communications device from data transport processing. The method includes the steps of processing requests to reserve bandwidth for a session of data communications and labeling a percentage of available data storage locations in the data communications device with a session identifier. Also, the method includes the step of concurrently processing and transporting data through a data communications device using the available data storage locations to store the data as it is processed, and depositing only data having a corresponding identifier equivalent to the session identifier of the storage locations into the data storage locations labeled with the session identifier. Using such a method, the device can reserve bandwidth while concurrently processing session data in the device.
In another method of the invention, the step of processing requests, processes requests to change an amount of reserved bandwidth associated with the session of data communication.
In yet another method embodiment, a method of storing bandwidth reservation information is provide and includes the steps of accepting a bandwidth reservation request indicating an amount of bandwidth to reserve for a session of data communication. Then, the step of calculating a percentage of total device bandwidth to allocate to the session of data communication based upon the bandwidth reservation request is performed. This is then followed by the step of storing the percentage in a resource allocation table which is independently accessible by a flow labeler.
Another embodiment of the invention provides a method for dynamically adjusting reserved bandwidth in a data communications device while transporting a session of data communication within the device. This method embodiment includes the steps of establishing a first bandwidth reservation associated with a session of data communication in the data communications device. This may be done, for example, by accepting a first bandwidth reservation request indicating a first amount of bandwidth to reserve for the session of data communication in the data communications device and by labeling, with an identity of the session of data communication, a first percentage of available data storage locations used to store data transported through the data communications device thus establishing the first bandwidth reservation. The first percentage of storage locations labeled is generally based upon the first amount of bandwidth requested as indicated in the first bandwidth reservation request.
Preferably, after the step of accepting a first bandwidth reservation request, the step of establishing a first bandwidth reservation further includes the step of calculating and storing a first percentage of total device bandwidth to allocate to the session of data communication based upon the first bandwidth reservation request. The first percentage of data storage locations labeled in the step of labeling is based upon the calculated first percentage of total device bandwidth to allocate to the session of data communication. Also, the step of calculating and storing preferably stores the calculated first percentage in a resource allocation table which is independently accessible by the step of labeling and the step of dynamically adjusting, so as to allow the step of dynamically adjusting to alter the calculated percentage in the resource allocation table without disrupting the step of labeling, thus allowing the bandwidth reservation in the device to be adjusted without effecting operation of a step of transporting (summarized below). Accordingly, data storage locations are labeled in accordance with the bandwidth requests and the labeling of the locations inherently reserves the bandwidth for sessions associated with the label.
As noted above, the embodiment also includes the step of transporting, through the data communication device, data associated with the session of data communication utilizing data storage locations associated with the first bandwidth reservation. The step of transporting can deposit the data associated with the session of data communication into data storage locations having an identification associated with the session of data communication and does so independently of how the identification associated with the session of data communication is created. Preferably, this step of transporting deposits the data associated with the session of data communication only into those data storage locations having an identification associated with the session of data communication. In other words, storage locations are labeled with an identity of a session of data communication for which bandwidth is reserved and during data transport, data associated with that session of data communication is placed into the labeled storage locations corresponding to the session. Preferably, other data not associated with the session of communication does not use the labeled locations, since they are reserved for the session data only. In this manner, the number of labeled locations selected from a total set of available location, such as labeling selected data storage locations in a single large rotating queue, are reserved for the session data.
The embodiment also includes the step of receiving bandwidth allocation adjustment information during the session of data communication. Preferably, this is done via a resource allocation protocol such as RSVP. That is, the data communications device uses an RSVP protocol to determine an amount of bandwidth to reserve.
Furthermore, the embodiment includes the step of dynamically adjusting the first bandwidth reservation to produce a second bandwidth reservation for the session of data communication in accordance with the bandwidth allocation adjustment information while continually maintaining the session of data communication.
In another embodiment based on the former embodiment, the step of dynamically adjusting the first bandwidth reservation to produce a second bandwidth reservation includes the steps of accepting a second bandwidth reservation request indicating a second amount of bandwidth to reserve for the session of data communication, and labeling, with an identity of the session of data communication, a second percentage of available data storage locations used to store data transported through the data communications device thus establishing the second bandwidth reservation. The second percentage of storage locations labeled is based upon the second amount of bandwidth requested as indicated in the second bandwidth reservation request, and the second percentage of storage locations labeled is different than the first percentage of storage locations labeled. This allows bandwidth to be adjusted by altering the labeled percentages for storage locations (e.g. in the single rotating queuing structure) associated with (i.e., labeled to receive) various sessions of data communication.
In another embodiment, after the step of dynamically adjusting the first bandwidth reservation to produce a second bandwidth reservation completed, the method further includes the step of calculating and storing a second percentage of total device bandwidth to allocate to the session of data communication based upon the second bandwidth reservation request. The second percentage of data storage locations labeled in the step of labeling is based upon the calculated second percentage of total device bandwidth to allocate to the session of data communication. The second percentage replaces the first percentage calculated previously.
Preferably, the step of calculating and storing stores the calculated second percentage in a resource allocation table as a replacement for the calculated first percentage. The step of calculating can include the steps of obtaining a current measurement of data communications device data storage locations available for data storage and a current bandwidth utilization rate and then computing an amount of bandwidth to reserve for the session of data communication based on the current bandwidth utilization rate and on the current measurement of data communication device data storage locations available for data storage.
The resource allocation table is independently accessible by the step of labeling and the step of dynamically adjusting, so as to allow the step of dynamically adjusting to alter the calculated first percentage in the resource allocation table without disrupting the step of labeling, thus allowing the first bandwidth reservation in the device to be adjusted without effecting operation of the step of transporting. The resource allocation table may be a database, table, linked list, object, or other data structure or storage mechanism used to store resource allocation data as described herein.
In another embodiment, the step of dynamically adjusting the first bandwidth reservation to produce a second bandwidth reservation includes the steps of accepting a bandwidth reservation request indicating a specific amount of bandwidth to reserve for the session of data communication. Next, a step of calculating and storing a percentage of total device bandwidth to allocate to the session of data communication based upon the bandwidth reservation request is performed, followed by a step of labeling, with an identity of the session of data communication, a percentage of available data communication device data storage locations used to store data transported through the data communications device. In this embodiment, the labeled percentage is based upon the calculated percentage of total device bandwidth to allocate to the session of data communication. Thus, the data storage locations are labeled according to reserved bandwidth requirements.
Other apparatus embodiments include computer program product(s) having a computer-readable medium including computer program logic encoded thereon for allocating bandwidth in a data communications device. The computer program logic, when executed on one or more processing units with the data communications device, cause the processing unit(s) to perform any and all of the aforementioned methods steps. That is, since certain embodiments of the invention can be implemented in software, the computer program embodiments cover a disk or other computer readable medium encoded with instructions to execute the invention as a software program. The disks or other mediums themselves containing the code are actual embodiments of this invention.
The resource allocation information for bandwidth reservations is preferably stored in the resource allocation table. In one embodiment, a computer readable medium is provided that is encoded with a data structure. The data structure stores bandwidth allocation information. The bandwidth allocation information includes an identity of at least one session of data communication and a number representing a percentage of data storage locations to associate with the identity of the at least one session of data communication. The number representing the percentage of data storage locations to associate with the identity of the session of data communication is preferably a number indicating a number of labels to apply to data storage locations so as to reserve the data storage locations for the data associated with the at least one session of data communication. This data structure embodiment can be used to maintain bandwidth reservation information within a data communications device and can be dynamically changed to re-apportion bandwidth resources while the device concurrently and separately maintains the sessions of data communication.
BRIEF DESCRIPTION OF THE DRAWINGS The foregoing and other objects, features and advantages of the invention will be apparent from the following more particular description of preferred embodiments of the invention, as illustrated in the accompanying drawings in which like reference characters refer to the same parts throughout the different views. The drawings are not necessarily to scale, emphasis instead being placed upon illustrating the principles of the invention.
FIG. 1 illustrates a typical prior art implementation of the RSVP protocol used to reserve bandwidth within a data communications device.
FIG. 2 illustrates the structure of a prior art packet used to transfer data according to the RSVP protocol.
FIG. 3 illustrates a data communications networking environment using data communications devices configured to reserve bandwidth according to the invention.
FIG. 4 illustrates an internal architecture and data flow diagram of a data communications device configured according to one embodiment of the invention.
FIG. 5 illustrates a more detailed architecture and data flow diagram for a data communications device configured according to the invention.
FIG. 6A illustrates a resource allocation table created illustrating example data flow resource allocations according to an embodiment of the invention.
FIG. 6B illustrates a detailed view of a queue entry labeling arrangement according to one embodiment of the invention.
FIG. 7 illustrates bandwidth policy and admission control processing steps performed by a bandwidth reservation processor configured according to one embodiment of the invention.
FIG. 8A illustrates resource allocation calculation processing steps performed according to one embodiment of the invention.
FIG. 8B illustrates queue entry label processing steps performed by a bandwidth labeler configured according to one embodiment of the invention.
FIG. 9A illustrates how data storage locations can be labeled according to percentages of flow bandwidth per flow according to one embodiment of the invention.
FIG. 9B illustrates how data storage locations can be labeled according to another embodiment of the invention.
DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An brief overview of the invention will assist in understanding the discussion of detailed embodiments. Generally, the system of the invention allows a data communications device to dynamically reserve bandwidth and adjust bandwidth reservations for sessions of data communication without session disruption. The device can perform reservation and adjustment operations independently of sessions of data communication that are in progress (i.e. actively being transmitted) in the device and that may be using any currently reserved bandwidth resources. However, as the bandwidth is adjusted, the session(s) for which bandwidth adjustment is made are transported according to the new bandwidth reservation. That is, adjustments in reserved bandwidth for a session of data communication can be made without concerning or bothering the continual process of transporting data for that session, but the session data itself will be transported using the new bandwidth reservation as it is put in place, whether the new reservation is an increase or decrease in available bandwidth.
By having the device separate the operation of bandwidth allocation and adjustment from the continual operation of transporting data through the device, the device can perform bandwidth reservations and adjustments without disturbing the flow or sessions of data communication. The system can accept requests to allocate or reserve bandwidth in a data communications device using bandwidth reservation protocols such as RSVP. The reservation requests create sender state data that can be used to compute resource allocation data. The device uses the resource allocation data to associate labels with data storage locations in a data storage mechanism according to the required bandwidth reservations. A data scheduling apparatus, which is ignorant of particular sessions having specific amounts of reserved bandwidth, examines data and deposits data into data storage locations having an associated label corresponding to a session identification specified in the data, if any. This way, the device only places session data into the labeled storage locations reserved for that session data. If the data contains an unknown session identification (or none at all), the data scheduler deposits data into a data storage location that is unlabeled or that has an unreserved label. Thus the percentage of data storage locations labeled in a device for a session determines bandwidth reserved for the session. The data storage labeling scheme of the invention operates separately from the data scheduling, enqueuing and data dequeuing mechanisms to allow data sessions to continue without interruption during bandwidth adjustments.
FIG. 3 illustrates an example of acommunications network200 configured according to the invention. Thenetwork200 includesdata links202 which interconnect data communications devices201-A through201-E,network policy server150, and hosts210 (including hosts210-A1,210-A2 and210-A3). Thedata links202 allow communications to take place between the various components shown in the figure and can be any type of communication medium including physical network cables, wires, fiber optic links, any type of wireless transmission links or another communications medium. Though thenetwork200 is illustrated as a relatively small network for ease of description of the invention, the invention is applicable to networks of all sizes, including interconnected local area networks (LANs), wide area networks (WANs), intranets, extranets, and conglomerations of many networks, such as the Internet, for example.
As illustrated, thehosts210 are general purpose computer systems such as personal computers, mini-computers, mainframes or the like that exchange data, as will be explained, over thenetwork200. It is to be understood, however, that thehosts210 may be many different types of computing or data exchanging devices such as file servers, web-site servers, network-telephony devices, audio or video servers, and so forth and that the invention is not limited to application only in a computer network or only for data exchange between devices of a specific type.
Thedata communication devices201 provide the processing resources (routing and switching algorithms, queues, buffers, switching fabrics, data busses, backplanes, input and output ports, and so forth) to propagate data through thenetwork200 between thehosts210. Thedata communication devices201 may be any type of data processing device that can transfer, switch, route or otherwise direct or propagate data in a network. Possible examples ofdata communications devices201 are network access servers, routers, switches, hubs, bridges, gateways, proxy servers, firewalls, modem banks, concentrators, repeaters, and similar data transfer devices, or any combination thereof. Preferred embodiments of invention are implemented within thedata communications devices201 and allow eachdevice201 to dynamically reserve bandwidth to one or more sessions of data communication betweenhosts210 and allow the amount of bandwidth that is reserved by or to those sessions to be changed without disrupting the sessions that are using the reserved bandwidth or that require a change in the amount of reserved bandwidth in eachdata communications device201.
FIG. 3 is suitable for illustrating some example operations of embodiments of the invention which are helpful in understanding more detailed embodiments presented later. InFIG. 3, suppose, for example, that host210-A1 is a video server that serves a stream of video packets203 (the “A” video stream) across thecommunications network200 to recipient hosts210-A2 and210-A3 using a multicasting protocol. Furthermore, assume that in order for a receiving host210-A2,210-A3 to properly receive the “A”video stream203 with adequate quality, the hosts210-A2,210-A3 require an end-to-end network bandwidth of 100 Kilobits per second (Kbps). That is, each data communications device201-B through201-E that transports the stream of “A”video packets203 between sending host210-A1 and recipient hosts210-A2 and210-A3 must supply a minimum data transfer rate (i.e., bandwidth) of 100 Kbps for the “A”video stream203.
Due to the critical or real-time nature of data in the “A”video stream203, the sending and receiving hosts210-A1,210-A2 and210-A3, in conjunction with the data communications devices201-B,201-C201-D and201-E use a bandwidth reservation protocol such as RSVP to establish and reserve a 100 Kbps channel for the “A”video stream203 through thenetwork200. Specifically, using RSVP, each data communications device201-B,201-C201-D and201-E receives RSVP path and bandwidth reservation request messages (not specifically shown in this figure) which specify, among other things, an identity of a specific session of data communication (the “A”video stream203 in this example), an amount of bandwidth to reserve for the session of data communication (100 Kbps in this example), and the path for which the requested bandwidth is to be reserved for the specified data stream (i.e.,203) within each particular device201-B through201-E. In an alternative embodiment, thenetwork policy server150 supplies the requests to each device201 (in the form of commands) which tell thedevices201 how much bandwidth to reserve for flows, streams, or sessions of data communication in thenetwork200.
According to the invention, eachdata communication device201 contains a bandwidth reservation processor500 (abbreviated B.R.P. inFIG. 4) and a data transporter300 (abbreviated D.T. inFIG. 4). In the illustrated example, only device201-B is illustrated with thebandwidth reservation processor500 and thedata transporter300, though it is assumed for this example that alldevices201 are configured in a similar manner. All of the processing associated with the reservation and allocation of bandwidth is performed by thebandwidth reservation processor500 in adevice201, while all of the processing associated with the transport of data (e.g., stream203) through adevice201 is handled separately and concurrently by thedata transporter300. Using this configuration, adata communications device201 can configure, control and adjust (if needed) bandwidth reservation requirements for streams of data (e.g., alter the 100 Kbps channel for the “A” video data stream203) independently of transferring the actual data (e.g., the “A” data stream packets) through thedevice201.
Continuing with the example, thebandwidth reservation processor500 in each device201-B through201-E receives the RSVP path and bandwidth reservation request messages. If thebandwidth reservation processor500 determines that a requesting application or host (e.g., receiving hosts210-A2 or210-A3) has permission or privileges to reserve the requested bandwidth (e.g., RSVP policy control) and also determines that the requested resource (e.g., the 100 Kbps bandwidth) is available in thedevice201, thebandwidth reservation processor500 in each data communications device201-B through201-E grants the request and establishes the 100 Kbps bandwidth reservation for the “A”data stream203 along the path from sending host210-A1 to receiving hosts210-A2 and210-A3. Once the bandwidth reservation is established, each data communications device201-B through201-E transports data (i.e., packets) associated with the session of data communication (i.e. the “A” video data stream203) using the reserved 100 Kbps resources (data storage locations in this invention, as will be explained).
Extending the example, assume that each recipient host210-A2 and210-A3 receives the “A”data stream203 at the reserved rate of 100 Kbps. That is, thebandwidth reservation processor500 configures each data communications device201-B through201-E with a bandwidth reservation of 100 Kbps of its total bandwidth (i.e., its total data transfer capacity or throughput for the path specified for, or required by, the data stream) for the “A”video stream packets203 which are continuously delivered to the recipient hosts210-A2 and210-A3 in real-time acrossnetwork200. If a video client application (not shown) executing on recipient host210-A3 senses that more network bandwidth is required (such as 120 Kbps) to effectively receive the “A”video data stream203, the host210-A3 can use RSVP to make a bandwidth reservation request (not shown) containing bandwidth allocation adjustment information to each network device201-E,201-D,201-C and201-B. The bandwidth allocation adjustment information in the bandwidth reservation request specifies a request for 120 Kbps of bandwidth to be reserved for the “A”video data stream203.
Using the invention, thebandwidth reservation processor500 in each device201-E through201-B along the path of the “A”data stream203 receives the RSVP bandwidth allocation adjustment information. Assuming bandwidth resources (i.e., an extra 20 Kbps) are available to meet the needs of the additional request (e.g. RSVP admission control), and that permission is granted for the requesting host (e.g.,210-A3) or client application to increase bandwidth to the requested level, thebandwidth reservation processor500 in each device201-E through201-B dynamically adjusts the original bandwidth reservation of 100 Kbps to produce a new bandwidth reservation of 120 Kbps for the “A”video data stream203 while continually maintaining (i.e., transporting) the “A”video data stream203. Essentially, the invention's implementation of the separation of bandwidth reservation, adjustment and control from the transportation of data through a data communications device, as configured according to the invention allows a session of data communication to be uninterrupted during adjustments to bandwidth for that session.
FIG. 4 illustrates a more detailed architecture of adata communications device201 configured according to one embodiment of the invention which provides the processing capabilities explained above. In this embodiment, thedata communications device201 contains thedata transporter300 including adata scheduler320 and adata storage mechanism340, and thebandwidth reservation processor500 including abandwidth request handler520 and abandwidth labeler550. At least oneinput port505 is provided in thedata communications device201 which is illustrated as receiving application or session data (e.g., the “A” videodata stream packets203 inFIG. 1) and RSVP reservation requests andpath messages511, shown as an “R” packet. Anoutput port506 is also provided which transmits data onto thenetwork200. Only one input andoutput port505,506 are illustrated for ease of description of the invention. It is to be understood that many ports serving as both input and output ports may exist in a preferred embodiment of thedata communications device201.
Thenetwork policy server150 is also shown in this embodiment to illustrate that thebandwidth reservation processor500 can receivecommands530 to govern bandwidth allocation operations (as explained herein), instead of usingbandwidth reservation requests511 fromindividual hosts201. This alternative arrangement may be beneficial when eachdata communications device201 network-wide is to have a permanent amount of dedicated reserved bandwidth for use by a special purpose application or network wide (e.g. multicast) session of data communication, for example.
According to the general operation of thedata communications device201, initial bandwidth reservation for a particular session of data communication is generally performed before data communication for the session actually begins. The invention however is equally applicable to situations where a session of data communication is already established (i.e., data transport is underway across the network) but there is no particular amount of bandwidth pre-allocated for that session in thedata communications devices201 which are transporting the data. That is, the invention can be used to establish a bandwidth reservation concurrently with an active session of data communication that is already being transported through a network without having to disrupt or interrupt the session in any way. In a similar manner, as explained in the above example, the invention can also be used to adjust or modify a bandwidth reservation already assigned to a data communication session that is underway and that is currently being transported through a network.
To reserve bandwidth for a session(s) of data communication in any of these situations, thedata communications device201 receives bandwidth reservation requests andpath messages511 which are directed to thebandwidth request handler520. Thebandwidth request handler520 is a process that executes on thedata communications device201 and is responsible for accepting or denying the bandwidth reservation requests511. If accepted, thebandwidth request handler520 provides one or more data structures called sender state data504 (FIG. 5) to thebandwidth labeler550.Sender state data504 specifies source and destination points for a particular session or sessions of data communication that exist (or that will exist) (i.e., the “A”video stream data203 inFIG. 3), a path (i.e., input port to output port) for the session or sessions of data communication, and an amount of bandwidth (e.g., 100 Kbps) required to be reserved for the session or sessions over the specified path.
The path in thesender state data504, in this example embodiment, indicates a route through which the session data travels (DATA IN, DATA and DATA OUT inFIG. 4) within thedata communication device201 from aninput port505 at which the session data is received, through the data transporter300 (to be explained shortly), to anoutput port506 which transmits the session data towards its destination. In this example embodiment, assuming a session of data communication enters and exits only through a single pair of input andoutput ports505,506 in thedata communications device201, the slowest port of a single input/output port pair505,506 limits the amount of bandwidth available. By way of example, assuming the input andoutput ports505,506 are configured the same (i.e., have the same maximum bandwidths), if theports505,506 each support a connection data rate of 400 Kbps, then thebandwidth reservation processor500 can reserve a maximum of 400 Kbps for a session of data communication on these ports. In the example provided above with respect to the initial bandwidth reservation provided for the “A”video data stream203 inFIG. 1, thebandwidth request handler520 producessender state data504 that specifies that a session of “A” video data requires 100 Kbps of bandwidth betweeninput port505 andoutput port506.
Once thebandwidth labeler550 obtains thesender state data504, thebandwidth labeler550 accesses512 thedata storage mechanism340 to establish the requested bandwidth reservation as specified in thesender state data504. Thebandwidth labeler550, as its name implies, operates (as will be explained in more detail) to label a certain percentage of data storage locations (not shown in this figure) maintained within thedata storage mechanism340 with the identity (i.e., a label) of the session of data communication for which bandwidth is to be reserved as specified in thesender state data504. Using the aforementioned example, thebandwidth labeler550 labels a certain percentage of data storage locations used to transfer data betweeninput port505 andoutput port506 in thedata storage mechanism340 with a label corresponding to the “A”video data stream203. The percentage of storage locations labeled is based upon the amount of bandwidth requested (100 Kbps) as indicated in the bandwidth reservation request, which is also provided in thesender state data504.
In this manner, bandwidth reservation is accomplished via use of thebandwidth reservation processor500 which accesses thedata storage mechanism340 to label certain data storage locations with a labels corresponding to the sessions of data communication requiring reserved bandwidth.
During a session of data communication, theinput port505 receives packets of application data203 (which in this invention generally refers to data transferred in a session of data communication) directs them to thedata scheduler320. The data scheduler320 schedules or deposits thedata packets203 into data storage locations (not specifically shown) within thedata storage mechanism340 which have corresponding labels provided by thebandwidth labeler550. Thedata storage mechanism340 then operates to transport theapplication data packets203 back onto thenetwork200 from anappropriate output port506, in order to send theapplication data203 further towards its eventual destination (e.g., one of receiving hosts210-A2,210-A3 inFIG. 1 in this example).
It is important to understand that thedata scheduler320 does not need to be made aware or provided with an indication of each different session (active or not) of data communication (i.e., “A” data stream203) for which bandwidth is reserved in thedata communications device201 configured according to this embodiment of the invention. Rather, thedata scheduler320 only needs to look at eachdata packet203 to determine if the packet is associated with any session of data communication and if so, thedata scheduler320 deposits the packet into a data storage location in thedata storage mechanism340 that has a corresponding label equivalent to the label for the session contained in the packet header180 (e.g., in theTspec field191 inFIG. 2). One ormore fields184 through191 inpacket header180, as explained with respect toFIG. 3, are preferably used to determine if aparticular packet203 is associated with any session(s) of data communication or not. In other embodiments, non-header fields such as data field183 (FIG. 2) may be used to determine if the packet510 (FIG. 2) (or cell, frame, etc.) is associated with a session of data communication corresponding to a labeled storage location.
In this manner, thebandwidth reservation processor500 reserves bandwidth in adata communications device201 without requiring thedata scheduler320 to be notified each time bandwidth is allocated for a session of data communication. This aspect of the invention also allows thebandwidth reservation processor500 to adjust bandwidth for a session of data communication without requiring any runtime changes or notifications to be made, or provided, to thedata scheduler320. That is, thedata scheduler320 can remain ignorant of how many sessions of data communication are currently active and/or have bandwidth reserved in thedata communications device201. Instead, thedata scheduler320 can focus on repetitively depositing packets into data storage locations having labels that match anRSVP packet header180, if any. If noRSVP packet header180 or other session identifier information exists for a packet, then thedata scheduler320 deposits that packet into any data storage location that is not presently labeled.
FIG. 5 illustrates a more detailed embodiment of adata communications device201 configured according to the invention. In the example illustrated inFIG. 5, there are four active flows ofdata203,204,205 and206 that are transported through thedata communications device201. Each flow is pictured at various positions of transit within thedevice201 as one or more small circular packets which are labeled with a letter “A”, “B”, “C” or “U” to indicate the flow to which that packet belongs. The flows “A”203, “B”204 and “C”205 represent sessions of data communication for which bandwidth is reserved in thedata communications device201. The flow “U”206, in which the “U” stands for unreserved, represents all other application data that is transported through thedevice201 for which there is no specific amount of bandwidth reserved. Since there is no specific bandwidth reservation established for the “U”data flow206, thedevice102 services the “U”flow206 in a best-effort manner using any remaining unreserved device bandwidth (e.g., using any unlabeled or unreserved data storage locations, such aslocation556 inFIG. 5).
In this embodiment, the data storage mechanism340 (FIG. 4) is represented as a single circular rotating queue structure340-1 that includes a number ofqueue entries345. In this particular example, there are twelveavailable queue entries345 arranged in the circular formation as shown inFIG. 5. Eachqueue entry345 is capable of storing one packet of data (i.e., one packet of aflow203,204,205 or a packet of unreserved data206) while the packet awaits transmission from a port (e.g. output port506) in thedevice201. In this illustration, the conveyor belt-like queue structure340-1 rotates in a clockwise direction. Thedata scheduler320 deposits data packets (e.g., packets fromflows203 through206) into thevarious queue entries345, as indicated by thearrows535, as will be explained in more detail shortly. Note that in this example, thedata scheduler320 can deposit data packets203-206 into more than onequeue entry345 at one time as they arrive at thedata scheduler320. This is indicated byarrows535 that point to more than onequeue entry345. As apacket203 through206 waits in the queue340-1, adequeuing mechanism350 services the queue340-1 at periodic intervals and thequeue340 rotates clockwise as thedequeuing mechanism500 services each entry345 (rotation indicated by the circular arrows at each end of the queue340-1) so that queue340-1 shifts the packets from left to right and closer to adequeuing mechanism350 on the right end of the queue340-1.
Thedequeuing mechanism350 removes thedata packets203 through206 from the queue340-1 as they appear at the right-most end and transfers the data packets from thedevice201 via output port(s)506 onto thenetwork200. The speed at which thedequeuing mechanism350 dequeues packets, the rotation of the queue340-1, and the number ofqueue entries345 that make up the queue340-1 generally determine the overall bandwidth that can be provided to transport data to theoutput port506. It is assumed in this example that the input andoutput ports505,506 can handle data faster that thedata transporter300 and that, for purposes of this explanation, thedata scheduler320 can examine the packet headers (e.g., RSVP header180) to determine where to direct a packet (e.g.,203 through206) at a rate that is greater than the overall maximum bandwidth for a port or session of data communication. Thus thedata scheduler320 does not act as a bottleneck in thedevice201.
Of particular importance to the invention is the manner in which thedata scheduler320deposits data packets203 through206 into the queue340-1 within thedata storage mechanism340. Since the invention in this embodiment eliminates the need for thedata scheduler320 to be made aware of what particular data flows or sessions of data communication (e.g.,203 through205) have associated reserved bandwidth at any point in time, thedata scheduler320 simply examines information in eachpacket203 through206 that arrives at theinput port505 and deposits that packet into aqueue entry345 that contains a label, such aslabel555 “C”, that matches the information examined in the packet (i.e.,203 through206).
The information examined in eachpacket203 through206 is preferably packet or RSVP header information contained in one or more ofheader fields180,181 and182, as illustrated inFIG. 2. As indicated above in the discussion ofFIG. 4, thebandwidth labeler550 labels (during the independent operation of the bandwidth reservation processor500) certain queue entries345 (i.e., data storage locations) based upon thesender state data504. This labeling process will be discussed in more detail shortly. In any event, aspackets203 through206 arrive from theinput port505, thedata scheduler320 determines if eachpacket203 through206 has an associated identification of one or more sessions of data communication (that may or may not have reserved bandwidth, which is unimportant as far as thedata scheduler320 is concerned), as specifically indicated, for example, by theTspec field191 in each packet510 (FIG. 2). Based on the value in theTspec field191, thedata scheduler320 deposits thepackets203 through205 (206 not having a Tspec field sincepacket206 is not associated with a particular flow or session of data communication having reserved bandwidth) intoqueue entries345 that have identification labels (e.g.,555) corresponding to the identification (Tspec field191) of the session of data communication for thosepackets203 through205.
In other words, thedata scheduler320 is coupled to theinput port505 to receivedata packet203 through206 associated with one or more sessions of data communication (e.g., “A”203, “B”204, etc.) and deposits thedata packets203 through205 associated with that session(s) into data storage locations (queue entries345 in this embodiment) associated with the bandwidth reservation for each session. Thedata scheduler320 deposits “U”Packets206 that are not associated with a particular session of data communication intoqueue entries345 that have either no label (i.e., do not contain alabel555 for a reserved bandwidth session) or a label indicating that theentry345 is unreserved, as illustrated by the example queue entry label “U”556.
In this manner, thedata transporter300 including thedata scheduler320 and thedata queuing mechanism340 operate independently of thebandwidth reservation processor500 to continually maintain and transport one or more sessions of data communication along with data (e.g.,206) not specifically associated with reserved bandwidth reserved in thedevice201. The operations of thedata scheduler320 and data storage mechanism340 (e.g. queue340-1) can be performed irrespective of the current bandwidth reservations (i.e., number of labeled queue entries345) that may exist or that may change for session(s) of data communication (e.g. streams “A”203, “B”204, “C”205).
Thebandwidth reservation processor500 including thebandwidth request handler520 and thebandwidth labeler550 operates asynchronously with thedata transporter300 mechanisms (e.g.,320,340) and is responsible for labeling thequeue entries345 in the queue340-1. Using the data storage location labeling techniques explained herein, bandwidth reservations are established and maintained for each session ofdata communication203 through205. The techniques also allow fordata206 which is not specifically associated with reserved bandwidth sessions to be transported as well.
More specifically, with respect to the example embodiment inFIG. 5, thebandwidth request handler520 receives bandwidth reservation requests and path messages511 (e.g., fromhosts210 onnetwork200 inFIG. 3). The reservation request andpath messages511, as previously explained, are used to request bandwidth reservations in thedevice201 for one or more flows or sessions of data communication, such as203 through205 in this example. In this embodiment, thebandwidth request handler520 includes abandwidth daemon501 which preferably is an RSVP protocol daemon process (e.g., RSVPD) that executes on a processor (not specifically shown) within thedevice201. Thebandwidth daemon501 receives the bandwidth request andpath messages511 and consults theadmission control module502 and thepolicy control module503 to determine, respectively, bandwidth resource availability and access control permission with respect to a requestinghost205 or application. Once access is granted and the bandwidth resources are determined to be available, thebandwidth daemon501 creates thesender state data504.
In this example embodiment,
sender state data504 includes, for each session of data communication for which bandwidth resources are requested (e.g., each session listed in the sender state data, such as
streams203 through
205), an identity of the session of data communication, an amount of bandwidth associated with the session of data communication, and the path (e.g., input/output port pair) through the
device201 that the session of data communication is to traverse using the reserved bandwidth resources. More specifically, Table 1 below illustrates an example of the
sender state data504 created by the
bandwidth daemon501, including some example requested bandwidth rates to be reserved for the sessions or flows “A”
203, “B”
204 and “C”
205 in
FIG. 5.
| TABLE 1 |
|
|
| Example ofSender State Data 504 |
| REQUESTED | |
| SESSION | RESERVED | SESSION PATH |
| IDENTIFICATION | BANDWIDTH | (PORT-TO-PORT) |
|
| “A” | 100 Kbps | Input Port 505-Output Port 506 |
| “B” | 64 Kbps | Input Port 505-Output Port 506 |
| “C” | 132 Kbps | Input Port 505-Output Port 506 |
|
As shown in Table 1, the request511 (FIG. 5) for the “A” session203 (i.e., hosts sending and receiving this stream or flow of data) indicates that 100 Kbps of bandwidth is to be reserved, while the “B”session204 has requested 64 Kbps of bandwidth, and the “C”session205 has requested 132 Kbps. In this example, each of these flows or sessions of data communication “A”202, “B”204 and “C”205 are traveling on the same path through thedevice201 configured with thesender state data504 in Table 1. In other words, in the example embodiment shown inFIG. 5, an assumption is made that queue340-1 services a single path within thedata communications device201. For example, queue340-1 may be associated with theoutput port506. Assuming there are many output ports in thedevice201, each output port (e.g.506) in this embodiment thus has its own associated queue structure similar to340-1 provided in order to buffer or store data packets (e.g.,packets203 through206) that are to be transported from that output port onto thenetwork200.
To assist in the explanation of the operation of this example embodiment, it is also assumed that the maximum total bandwidth for theoutput port506 is 400 Kbps. Thus when thedata scheduler320, queue340-1 anddequeuing mechanism350 all operate at peak capacity, a maximum bandwidth or throughput of 400 Kbps is available from theoutput port506.
To configure bandwidth reservations for each flow or session of data communication (e.g., “A”, “B”, “C”) defined in thesender state data504, thebandwidth labeler550 in this embodiment includes aresource allocation calculator552, a resource allocation table553, and alabel calculator554.
Theresource allocation calculator552 creates labeling information that is maintained in the resource allocation table553 based on thesender state data504. To do so, in this embodiment, theresource allocation calculator552 obtains asinput560 the size and speed of rotation of the data storage mechanism340 (e.g. queue340-1). Essentially, theresource allocation calculator553 calculates and stores a percentage of total bandwidth (for the path to theoutput port506 in this example) to allocate or reserve for each session of data communication based on thesender state data504 as defined by the received bandwidth reservation requests511.
FIG. 6A illustrates an example of the contents of the resource allocation table553. In Column1 “FLOW ID”, the resource allocation table553 provides an identification of each flow or session of data communication (e.g., “A”, “B”, “C”), and includes an entry marked “U” representing resource allocation data for all data (e.g. packets206) not specifically associated with any session or reserved bandwidth.Column2 “PERCENT UTILIZATION” indicates a percent utilization of total bandwidth for the queue340-1 to which this resource allocation table553 is associated.Column3 “ENTRY COUNT OF TOTAL QUEUE SIZE” indicates the number ofqueue entries345 that should be labeled in the queue340-1 for each particular flow (i.e., row) represented in the resource allocation table553.
As noted above, theresource allocation calculator552 obtains as input the queue size andspeed data560 from thedata storage mechanism340. Using this information in conjunction with thesender state data504, theresource allocation calculator552 computes PERCENT UTILIZATION and ENTRY COUNT OF TOTAL QUEUE SIZE values for each flow or session of data communication (e.g., “A”203, “B”204, “C”205) in thesender state data504.
Theresource allocation calculator552 computes PERCENT UTILIZATION by converting the requested bandwidth to be reserved for each data flow or session to a percentage of total bandwidth for the queue340-1. In this embodiment, the queue size/speed data560 determines total bandwidth for the path. In this example, the queue340-1 has a queue size equal to twelvequeue entries345. Eachentry345 can store one packet (e.g., onepacket203 through206). A packet size in this embodiment is assumed to constant at 1000 bits. The queue speed in this example is assumed to be 33⅓ rotations per second. Based on these values, the total bandwidth for this queuing structure340-1 can be computed as follows:
QUEUE_BANDWIDTH=TOTAL_NUMBER_OF_ENTRIES*ENTRY_SIZE*ROTATIONS_PER_SECOND.
Or for this particular embodiment, overall total queue bandwidth equals:
400Kbps=(12 Queue entries)*(1000 bit entry size)*(33⅓ Rotations per second).
Once the total queue bandwidth is calculated, PERCENT UTILIZATION may be calculated as follows:
PERCENT UTILIZATION=REQUESTED_BANDWIDTH_PER_FLOW/QUEUE_BANDWIDTH.
In the example embodiment, since the overall queue bandwidth is 400 Kbps using the requested reserved bandwidth values from Table 1 (sender state data
504), the PERCENT UTILIZATION for each FLOW ID for flows “A”, “B” and “C” (
203 through
205) is computed as follows:
| |
| |
| FLOW ID A | .25 = 100 Kbps/400 Kbps |
| FLOW ID B | .16 = 64 Kbps/400 Kbps |
| FLOW ID C | .33 = 132 Kbps/400 Kbps |
| Total Reserved: | .74 * 100 = 74 Percent. |
| |
Any remaining bandwidth, expressed as a percentage (i.e., 100 percent−Total Reserved percentage), is allocated to all unreserved data packets206 (e.g. all data with FLOW ID=“U”). In this example, data associated with FLOW ID “U” gets 26 percent of the total bandwidth, since that is the remaining amount of bandwidth not reserved to the sessions ofdata communication203 through205.
Theresource allocation calculator552 computes ENTRY COUNT OF TOTAL QUEUE SIZE (Column3) values in the resource allocation table553 based on the percent utilization of each FLOW ID in proportion to the number oftotal queue entries345 in the queue340-1. The final result may be rounded or the allocation can be adjusted over each rotation of the data storage mechanism340-1 so the average transmission rate converges to the correct percentage. Specifically, in this example embodiment:
ENTRY COUNT OF TOTAL QUEUE SIZE=PERCENT UTILIZATION*TOTAL_NUMBER_OF_ENTRIES.
In the example embodiment, since there are twelve
queue entries345 in the queue
340-
1, the ENTRY COUNT OF TOTAL QUEUE SIZE values are computed as follows:
| |
| |
| FLOW ID A | 3 = .25 * 12 |
| FLOW ID B | 2 = .16 * 12 |
| FLOW ID C | 4 = .33 * 12 |
| FLOW ID U | 3 = .25 * 12 |
| TOTAL LABELED | 12 |
| |
Once theresource allocation calculator552 computes the data in the resource allocation table553, thelabel calculator554 can use the resource allocation table553 to properly produce labels (e.g.,555) for thequeue entries345 in the queue340-1. As indicated inFIG. 5, thelabel calculator554 accesses the resource allocation table553 to determine howmany entries345 are to be labeled with each particular session identification label or FLOW ID which identifies the session (FIG. 6A). As indicated in the example resource allocation table553 inFIG. 6A, thelabel calculator552 labels threequeue entries345 with “A”, twoqueue entries345 with “B”, fourqueue entries345 with “C”, and threequeue entries345 with “U”, for a total of twelve labeled queue entries that make up the total queue340-1.
FIG. 6B illustrates in more detail an example structure of thequeue entries345 that make up the queue340-1. Each entry345-1,345-2 and345-3 is essentially a data storage location346-1,346-2,346-3 that includes an associated label portion555-1,555-2 and555-3, respectively. The data scheduler321 deposits the data (i.e. packets310) into the datastorage location portion346 of theentry345 in this figure. As indicated in this sample segment of thequeue340, the data storage location346-1 for queue entry345-1 currently stores an A-PACKET310-1 that is associated with the “A” session of data communication203 (FIG. 5), while the data storage location346-2 for queue entry345-2 stores a B-PACKET310-2 that is associated with the “B” session of data communication204 (FIG. 5), and the data storage location346-3 for queue entry345-3 stores a C PACKET310-3 that is associated with the “C” session ofdata communication205. These packets310-1,310-2 and310-3 are stored in these respective locations346-1,346-2 and346-3 because the labels for these locations555-1,555-2 and555-3 created by thelabel calculator554 in thebandwidth labeler550 match the session or flow identification information contained in theheader180 of each packet310-1,310-2 and310-3. That is, thedata scheduler320 places the A-packet310-1 into the location346-1 because the label555-1 indicates that the location346-1 is reserved for “A”data203.
Accordingly, by labeling thedata storage locations345 that form the queue340-1 with theappropriate labels555 for each flow identification (FLOW ID, Column1) specified in the resource allocation table553, thebandwidth labeler550 can constantly maintain the appropriate amount of reserved bandwidth for each session ofdata communication203,204,205. Thedata scheduler320 uses theentries345 that are either unlabeled by thebandwidth labeler550 or that are labeled with a “U” (signifying unreserved or unlabeled), as indicated bylabel556 inFIG. 5, to deposit any data (e.g. data packets206) that does not contain a flow or session identification.
In this embodiment, thebandwidth labeler550 can continuously monitor thesender state data504 for changes in bandwidth requests (i.e., bandwidth reservations or changes that have been granted by the bandwidth daemon501) for any session of data communication (e.g.203,204,205). Once a change is detected in thesender state data504, theresource allocation calculator552 recalculates the values in the resource allocation table553. Thelabel calculator554 detects this change and then correspondingly alters the labeling of thequeue entries345 to effectuate the requested bandwidth change.
In this manner, the system of the invention allows bandwidth to be dynamically adjusted without affecting thedata scheduler320. That is, since thebandwidth reservation processor500 adjusts a proportionate number oflabels555,556, etc. for sessions of data communication within thequeue entries345, the maximum allowable bandwidth for the sessions (e.g.203,204 or205 in this example) is inherently governed, since thedata scheduler320 can only place as many packets of session data (i.e.310) into labeledqueue entries345 as there are matchinglabels555 associated with theentries345.
By isolating the operation of thedata scheduler320 from thebandwidth reservation processor500 as shown in the previous embodiments, it has been illustrated how bandwidth may be reserved and adjusted dynamically at any time before, during, or after one or more sessions of data communication (e.g.203 through205) are in operation. Once thedata scheduler320 queues a data packet310 for a session of data communication in the queue340-1, the data310 remains in thedata storage location346 until it is dequeued by dequeuingmechanism350. Thelabel calculator554 in this embodiment only labels555 or changes labels ofqueue entries345 that do not already contain data310. In this manner, thedata scheduler320 and queue340-1 operate in continuously the same manner. This allows any session data in the “pipeline” comprising theinput port505,data scheduler320, queue340-1,dequeuing mechanism350, andoutput port506 to remain undisturbed during a change in bandwidth.
Recall that prior art implementations of bandwidth reservation require a session of data communication to be broken or halted for a period of time while classification, scheduling, queuing, and dequeuing mechanism are all reconfigured to handle the new bandwidth requirements. Once reconfigured, the session can then be reinstated. The system of the invention avoids much of this effort and allows the session to be continually transmitted before, during and after bandwidth allocations or adjustments. The adjustments take effect dynamically as the new label configurations for thedata storage locations345 in queue340-1 are used. Thus thebandwidth labeler550 can dynamically re-labelqueue entries345 and the new labels are used by thedata scheduler320 to deposit session data.
An example highlights the particular importance of this aspect of the invention. Suppose that thedevice201 is currently transporting the “B”session204 of data communication at a maximum bandwidth of 64 Kbps, as illustrated inFIG. 5 and in the resource allocation table in553 inFIG. 6A. Next assume that the “B”stream204 requires more bandwidth. The new bandwidth required might be, for example, 100 Kbps. This may be sensed by one of the receiving hosts205-A2,205-A3 (FIG. 3), for example, that determines that the current allocation of 64 Kbps is insufficient and that an additional 36 Kbps would correct the situation. Using RSVP or another bandwidth reservation protocol, thebandwidth request handler520 receives the request for additional bandwidth511 (either a request for an additional 36 Kbps, or a request for a change from 64 Kbps to 100 Kbps, or a new reservation request for 100 Kbps for the “B” stream).
Assuming thebandwidth daemon501 grants therequest511, thesender state data504 for the SESSION IDENTIFICATION “B” indicates a REQUESTED RESERVED BANDWDITH value of 100 Kbps. Theresource allocation calculator552 detects the change in thesender state data504 and updates the resource allocation table553 as explained above so that FLOW ID “B,” (which was set at 16 percent with 2 queue entries labeled with “B”), now contains a PERCENT UTILIZATION value of 0.25 (or 25 percent of the total 400 Kbps bandwidth for this path) and an ENTRY COUNT OF TOTAL QUEUE SIZE value of 3. When the resource allocation table553 is updated in this manner, thelabel calculator554 detects the change and begins to re-label thequeue entries345 according to the new information in the resource allocation table553. Once thelabel calculator554 re-labels allentries345 in the queue340-1, threeentries345 are labeled with a “B” instead of two as in the previous configuration. Note that thelabel calculator554 preferably operates to re-labelqueue entries345 just after thedequeuing mechanism350 removes the data (310 inFIG. 6B) from eachentry345. The “X” in thedata storage location346 inFIG. 5 indicates that thequeue entry346 is now void of any data packet310 and can be re-labeled if required.
Alternative embodiments of the invention provide that thelabel calculator554 always operates to continuously labelentries345 according to the resource allocation table553. In this manner, if the number of labels required for all sessions having reserved data exceeds the total number ofqueue entries345 in the overall queue340-1, eachentry345 is provided with a different label upon being emptied by thedequeuing mechanism350. In other words, if therotation speed560 of a short queue340-1 (i.e. a queue340-1 having sofew entries345 that allentries345 combined cannot hold the total amount of reserved bandwidth) is fast enough, thelabel calculator554 can simply provide labels for everyentry345 after that entry passes thedequeuing mechanism350. In this manner, a short queue changes its labeling configuration with each rotation, and the label calculator controls the bandwidth allocation for each session via the labels for theentries345.
All data packets (e.g., packets forsessions203 through205 and unreserved data packets206) that currently exist in the queue340-1 during the labeling process remain queued and eventually propagate their way to thedequeuing mechanism350. Preferably, relabeling takes place as soon as eachentry345 in the queue340-1 is emptied or dequeued of its data packet310 by thedequeuing mechanism350. As the relabeledqueue entries345 make their way clockwise around to thedata scheduler320 to obtain more data packets310, the new labeling configuration (i.e., the queue340-1 now containing three “B” labeled entries345) will dictate what data can be placed into whichentries345. In this manner, the bandwidth can be changed for the session ofdata communication204 without disrupting the transport of data for the session.
FIG. 7 illustrates the processing steps performed by thebandwidth request handler520 configured according to this invention. Instep600, thebandwidth allocation request511 is obtained from thenetwork200. Instep601, thebandwidth daemon501 determines the requested resource availability via admission control. If the requested resource is not available, thebandwidth daemon501 processing denies the request and returns to step600. If thebandwidth daemon501 instep601 does not deny the request, thebandwidth daemon501 instep602 authenticates access to the requested resource via policy control. If instep602 thebandwidth daemon501 determines that the access should not be granted to the requested resource, thebandwidth daemon501 denies the request and processing returns to step600. Ifsteps601 and602 pass,step602 directs processing depending upon the type of request (511) received. If therequest511 is a new request to reserve bandwidth for a session that does not yet have bandwidth reserved, the processing proceeds to step603 and thebandwidth daemon501 produces the newsender state data504 for a session identification associated with the newly requested resource. If however step602 determines that therequest511 is requesting alteration of a resource already reserved to a particular session or sessions of data communication, then processing follows as explained instep604.
Instep604, thebandwidth daemon501 updates thesender state data504 that already exists for the requested resource, without disturbing any session or sessions of data communication that may be using that resource (i.e., without notifying the data scheduler320). Instep605, thebandwidth daemon501 makes thesender state data504 available to thebandwidth labeler550 so that thebandwidth labeler550 can label (e.g.,555) the data storage locations (e.g., entries345) accordingly in thedata storage mechanism340, which is preferably the rotating queue structure340-1 discussed above. By making thesender state data504 available to thebandwidth labeler550, thebandwidth request handler520 can focus its operation primarily on bandwidth request processing and does not need to make thesender state data504 available to other components of the system, such as thedata scheduler320 or thedequeuing mechanism350.
FIG. 8A shows the general processing steps performed by theresource allocation calculator552 in thebandwidth labeler550 according to one embodiment of the invention. Instep700, theresource allocation calculator552 queries thesender state database504. Alternatively, step700 may be performed by having any changes in thesender state data504 be signaled (i.e., viastep605 inFIG. 7) to theresource allocation calculator552. Instep701, current statistics of thedata storage mechanism340 such as queue size (e.g., total number of entries345) and speed (e.g., how many entries are dequeued over a period of time, rotation, etc.) are queried to determine the overall current bandwidth characteristics for the requested path. Step702 then calculates and/or updates the resource allocation table553 values (Columns1,2 or3) with the current session attributes, as explained above. In this manner, the processing ofFIG. 8A converts thesender state data504 into meaningful data usable by thelabel calculator554 to label (555 inFIG. 6B) the data storage location346 (i.e., queue entries345) in the data storage mechanism340 (e.g., queue340-1).
FIG. 8B shows the general processing steps associated with thelabel calculator554. Instep750, thelabel calculator554 queries the resource allocation table553 for flow label identification (i.e., Columns1 and3). This step could be triggered by a change in the resource allocation table553, or may be performed periodically or continuously. Step751 then determinesqueue entry345 label allocations.
In one embodiment,step751 consults the entry label counts for each session of data communication as indicated inColumn3 of the resource allocation table553. Step752 then labels555 theentries345 according to the entry label calculations. The labeling ofqueue entries345 may proceed serially by labelingentries345 with all of the “A” labels (e.g., 3 “A” labels555), and then when there are no more “A” labels remaining,labeling entries345 with “B” labels (2 in the example) until none remain, and so forth. With respect to the example resource allocation table553 inFIG. 6A, the twelve labeled entries in the example queue340-1 inFIG. 5 would have a labeling order as follows:
- “A” “A” “A”“B” “B” “C” “C” “C” “C” “U” “U” “U”
However, as will be explained next with respect toFIGS. 9A and 9B, thelabel calculator554 may label theentries345 in thequeue340 instep752 in a variety of other patterns so as to evenly distribute labeledentries345 for each session or FLOW ID, depending upon the selected embodiment.
FIG. 9A illustrates an example of a labeling pattern. InFIG. 9A, thebandwidth labeler550 sequentially “uses up” the labels for each session having a reserved resource in a serial manner, and when none are left, moves on to the next set of labels. As indicated in the figure, which corresponds to the bandwidth reservations established in the resource allocation table553 inFIG. 6A, the “A” session ofdata communication203 has twenty-five percent of the bandwidth reserved via “A” labels in thequeue entries345. Thus thedata scheduler320 inFIG. 5 is able to queue “A”data packets203 into one quarter of the entire queue space on each rotation. The “C” data stream has thirty-three percent of the queue reserved, and the “B” stream has sixteen percent reserved. This leaves a remaining twenty-five percent of thequeue entries345 labeled with “U”, or not labeled at all. The “U” labeledentries345 are used for alldata206 that does not belong to a session having a bandwidth reservation in thisdevice201.
It may be apparent to those skilled in the art that a situation might arise where thedata scheduler320 detects an RSVP header that indicates a session identification (i.e., a labeled packet) for which there are currently no corresponding labeledqueue entries345. The invention addresses this situation in a number of ways. First, thedata scheduler320 can simply buffer the data until anentry345 having acorresponding label555 appears. After a certain time period, which preferably corresponds to a certain number of rotations of thequeue340, if no labeledentry345 appears (to the data scheduler320) that matches the packet data (with a session identification) that is buffered with the unknown session identification, thedata scheduler320 can either discard the unknown session data or can simply deposit the data into one of the data storage locations that is indicated as being unreserved (i.e. labeled “U”). The later mechanism (queuing into an unreserved entry345) is preferred over the packet discard mechanism, since data will not be lost and will not require re-transmission from the sender if the unknown data stream has such error detection/correction capabilities enabled.
In this manner, if session data is transported to adevice201 which is unaware of the existence of the session, the invention still allows the data to be transported as if it were unlabeled data not associated with any particular session. In prior art systems in which a classifier and scheduler are made aware of all active sessions, the unknown session data might confuse the classifier and/or scheduler and may require either, at a minimum, to pause operation to consult with the RSVP daemon (e.g.,101 inFIG. 1) to determine how to handle the unknown session data. The invention avoids such cumbersome approaches and keeps the data transport mechanism separated from the bandwidth allocation and administration aspects of thedevice201.
FIG. 9B illustrates another labeling pattern which can be used by thelabel calculator554 to labelqueue entries345 with session labels. The approach taught inFIG. 9B is cycling label approach. In this approach, thelabel calculator554 repetitively cycles through each FLOW ID in the resource allocation table553 and labels onequeue entry345 for each session or flow id per cycle. During the repetitive cycling, thelabel calculator554 decrements the number of labels remaining for each FLOW ID. When a FLOW ID has no labels remaining (i.e., its ENTRY COUNT OF TOTAL QUEUE SIZE value is zero), no more labels are created for that session or flow identification. In this manner, a more balanced approach is provided for the queue entry labeling process of the invention, since each flow is provided with a labeledentry345 that is separated from another similarly labeledentry345 by other entries labeled for other flows that still have more bandwidth to be reserved (i.e., more entries that are to be labeled). As illustrated inFIG. 9B, from left to right, the labels onqueue entries345 appear in this example as follows:
- “A” “B” “C” “U” “A” “B” “C” “U” “A” “C” “U” “C”
In total, there are three “A” entries, two “B” entries, four “C” entries, and three “U” entries, and that the entries are somewhat staggered from each other. The sequence A-B-C-U begins at the left and repeats itself twice, after which there are no more “B” labels to be produced, and so the remaining “A” and “U” labels are produced. In this manner, thebandwidth labeler550 presents a more even distribution of labeled (i.e., reserved)bandwidth entries345 so that thedata scheduler320 does not have to wait for significant periods of time while buffering data and awaiting for an entry with the correct label to appear.
In yet another embodiment of thebandwidth labeler550 and thelabel calculator554, thelabel calculator554 only labels thequeue entries345 each time the resource allocation table553 changes. As such, the labels (e.g.555 inFIGS. 5 and 6B) remain allocated or associated with eachentry345 as theentry345 continually circulates around and around thequeue340. In this manner, the bandwidth reservations for each session are static until they need to be changed. That is, the only time thequeue entries345 are relabeled is if a change is detected to the bandwidth reservations as communicated by the change in data values in the resource allocation table553. This embodiment conserves processing resources used by thebandwidth labeler550, which can enter an idle state until thesender state data504 changes. The change causes theresource allocation calculator552 to “wake up” and update the resource allocation table553, which in turn causes thelabel calculator554 to re-labelentries345 as required.
It is to be understood by those skilled in the art that the labeling patterns inFIGS. 9A and 9B are illustrative as examples only, and are not limiting of the present invention. Rather, other fair, weighted, or even distribution schemes known to those skilled can be used to label the sequence ofqueue entries345 so as to best distribute the reserved bandwidth for each session of data communication across the entire queue340-1. For example, in an alternative embodiment, eachqueue entry345 may be larger than the size of a single packet. In such cases, anentry345 may hold many packets, cells, frames, or other unit of data from the session of data communication. In another alternative embodiment, each entry may have more than one label. That is, is two or more sessions of data communication are somehow related, or have equivalent bandwidth reservations (e.g., same percentage for both sessions), the bandwidth reservation processor might label a single entry with more than one session identification. In this manner, thedata scheduler320 can deposit any one or a number of different packets into themulti-labeled entry345.
It is also to be understood that the invention is not limited to applications providing bandwidth reservation and allocation using the RSVP protocol. Rather, the invention is intended to operate in conjunction with other bandwidth reservation, allocation, or adjustment protocols that currently exist or that may be developed in the future. For example, future versions of RSVP may provide specific message formats to enable bandwidth adjustments. The invention provides implementations of data communications devices as explained herein that can take advantage of such messages to dynamically adjust bandwidth as required for sessions of data communication.
For more details on the operation of bandwidth reservation protocols such as RSVP and its derivatives, the reader is directed to Request For Comments 2205 and 1633 and RSVP93 (RFC-2205, RFC-1633, RSVP93), published by the Network Working Group of the Internet Engineering Task Force (IEFT), and available on the Internet at ftp://ftp.isi.edu/in-notes/rfc2205.txt, each of which protocol references is hereby incorporated by reference in their entirety.
The invention applies to all types of data transmitted to or from any type of device through any type of network and/or network communications medium. While the illustrated examples discuss packet data which is primarily applicable to Transmission Control Protocol/Internet Protocol (TCP/IP) networks such as the Internet, the invention is equally applicable to networks that use such units of data as tokens, cells, frames, blocks, and so forth. Other network architectures such as Asynchronous Transfer Mode (ATM) networks can use the concepts of the invention as well to reserve bandwidth for cell transfer. Also, networking architectures such as packet-wireless, Fiber Distributed Data Interface (FDDI), Systems Network Architecture (SNA), Digital Subscriber Link (DSL), Advanced Peer-to-Peer Networking (APPN) and others may benefit from use of the invention.
Another alternative scenario that could illustrate the features of the invention would be to have several networked computers each running different types of applications having different data communications requirements. The data produced from each application may need to be transferred between the computers at different reserved rates. The invention could be used to provide this capability. It is also understood that adata communications device201 configured according to the invention may have one ormore data schedulers320 and one ormore data queues340. An arrangement such as a single data scheduler per input port that can deposit data into many different queues340-1,340-2, etc., where there is onequeue340 per output port is contemplated as a device configured according to the invention. Other arrangements are possible as well which are contemplated by the invention. Such alternative arrangements and alternative designs of data communications devices can apply the concepts of the invention as disclosed herein to provide dynamic bandwidth reallocation without interrupting streams of data, since the operation of the bandwidth allocation mechanisms are generally separated from the data transport mechanisms, as explained herein.
While this invention has been particularly shown and described with references to preferred embodiments thereof, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.