Movatterモバイル変換


[0]ホーム

URL:


CN109240832A - A kind of hardware reconstruction system and method - Google Patents

A kind of hardware reconstruction system and method
Download PDF

Info

Publication number
CN109240832A
CN109240832ACN201811113981.5ACN201811113981ACN109240832ACN 109240832 ACN109240832 ACN 109240832ACN 201811113981 ACN201811113981 ACN 201811113981ACN 109240832 ACN109240832 ACN 109240832A
Authority
CN
China
Prior art keywords
pcie
crossbar
resource pool
connection
virtual channel
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201811113981.5A
Other languages
Chinese (zh)
Other versions
CN109240832B (en
Inventor
袁柳
魏星
涂吉
谢海永
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
China Electronics Technology Group Corp CETC
Original Assignee
China Electronics Technology Group Corp CETC
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by China Electronics Technology Group Corp CETCfiledCriticalChina Electronics Technology Group Corp CETC
Priority to CN201811113981.5ApriorityCriticalpatent/CN109240832B/en
Publication of CN109240832ApublicationCriticalpatent/CN109240832A/en
Application grantedgrantedCritical
Publication of CN109240832BpublicationCriticalpatent/CN109240832B/en
Activelegal-statusCriticalCurrent
Anticipated expirationlegal-statusCritical

Links

Classifications

Landscapes

Abstract

Translated fromChinese

本发明公开了一种硬件重构系统及方法,本发明可以通过PCIe多总线和交叉开关技术将GPU、FPGA、GPU、IPU、SSD等计算和存储资源从服务器中解耦出来,并实现异构计算和存储资源池的池内互联、池间互联等层次化的互联。通过基于池内和远程PCIe交叉开关的配置提供标准PCIe计算资源的连接重构能力。通过PCIe交换机制的引入,充分利用PCIe通信的可扩展性及性能优势,可以创建性价比高的高可用性系统,可支持大部分异构设备,通用性更强,并且可以减少数据传输延迟、系统复杂性和功耗。

The invention discloses a hardware reconstruction system and method. The invention can decouple computing and storage resources such as GPU, FPGA, GPU, IPU, and SSD from the server through PCIe multi-bus and cross-switch technology, and realize heterogeneous Hierarchical interconnections such as intra-pool interconnection and inter-pool interconnection of computing and storage resource pools. Provides connection reconfiguration capabilities for standard PCIe computing resources through in-pool and remote PCIe crossbar-based configurations. Through the introduction of the PCIe switching mechanism, the scalability and performance advantages of PCIe communication can be fully utilized, and a cost-effective high-availability system can be created, which can support most heterogeneous devices, has stronger versatility, and can reduce data transmission delay and system complexity. performance and power consumption.

Description

A kind of hardware reconstruction system and method
Technical field
The present invention relates to field of computer technology, more particularly to a kind of hardware reconstruction system and method.
Background technique
At present in magnanimity node server use process, each component resource of presence server (CPU, memory, disk,SSD, network I/O, GPU, FPGA etc.) utilization rate it is unbalanced caused by the wasting of resources and energy waste and these component resourcesHigh cost caused by respectively server hardware caused by upgrading period difference frequently upgrades.Domestic data center serviceMostly by Ethernet connection there are agreements too fat to move, the high problem of power consumption between each hardware resource of device, and InfiniBand etc. is evenSpecial equipment is connect and needs, device dependence is strong.The enterprises such as Baidu, Tencent, Huawei are started to be calculated with server, storage, be transmittedThe decoupling of equal components resource and pond turn to direction and carry out architectural studies, including pond storage service, Heterogeneous Computing resource (FPGA,GPU, CPU, IPU) pondization service etc..But it is limited only to complete machine level currently based on PCIe hardware resource pool, and also visitingRope does not form standard.And merely with PCIe extended capability, it is ignored in the advantage for communicating and reconstructing scheduling.
Summary of the invention
The present invention provides a kind of hardware reconstruction system and methods, to solve in prior art hardware reconstruction to all partsThe unbalanced problem of utilization rate.
On the one hand, the present invention provides a kind of hardware reconstruction systems includes:
Between each resource pool and the resource pool closed on, the multichannel interface connection based on remote PCI e crossbar switch, resourceMultichannel interface connection in pond based on PCIe crossbar switch, remote PCI e crossbar switch and the PCIe crossbar switch in resource pool,It is connected together by shared interface, forms the double-deck crossbar configuration based on PCIe crossbar switch and remote PCI e crossbar switchLink;
Wherein, the resource pool includes computing resource pool and/or memory resource pool, and each processor passes through resource in resource poolPCIe crossbar switch inside pond is interconnected, and when being attached configuration, is configured inside resource pool by configuration informationPCIe crossbar switch connection relationship is calculating and storage resource inside the specified PCIe bus assignment resource pool of CPU, realizes meterIt calculates and is directly connected between resource pool, memory resource pool inner treater, support computing resource and the distribution of storage resource dynamic on-demand;
The remote PCI e crossbar switch is interconnected by shared interface and each resource pool, realizes the resource based on PCIe protocolPoint-to-point peer switch between pond configures the connection of remote PCI e crossbar switch by configuration information when being attached configurationRelationship is the processor resource in the specified PCIe bus assignment Adjacent resource pond of CPU, carries out calculating and storage money between resource poolSource dynamically distributes.
Preferably, the shared interface supports two kinds of interface modes of transparent bridge and non-transparent bridge, can be configured to main sideMouthful or from port, interconnected between realizing port by PCIe shared interface cable.
Preferably, the PCIe crossbar switch in the resource pool includes with the realization of PCIe switching backplane, PCIe switching backplaneMultichannel PCIe bus slot and PCIe exchange chip.
Preferably, the remote PCI e crossbar switch is realized with PCIe switch, realizes the point-to-point exchange between resource poolMode, including one of following or a variety of: packet switch, circuit switching and virtual channel exchange.
Preferably, virtual channel exchange multiple shares the same physical channel for being made up of virtual channelVirtual circuit exchange connection, and virtual circuit exchange connection can connect with packet switch connection and circuit switching cooperate to transmitFlow data.
It preferably, further include compiler;The compiler is used for according to known internuclear stream communication situation, for everyKnow stream communication distribution path, and determine that virtual circuit exchange connection, circuit switching connection are connected with packet switch, at runtime basisAs a result, establishing each exchange connection by the link information prestored, the communication between different resource pond is respectively being corresponded to for communication compilingExchange connection on transmitted.
Preferably, the remote PCI e crossbar switch includes multiple input units, and a crossbar switch crosses over unit, pathComputing unit, virtual channel allocation unit, crossbar switch allocation unit and circuit configuration unit;
The input unit, for the virtual channel number after the PCIe device configuration output arbitration for connection, each inputUnit includes n output virtual channel VC1-VCn, and a bypass channel, a PS state stores and a VCS state storage,The input unit exchanges VCS signal with the input virtual channel of PCIe device and port input signal is connected, the storage of PS stateThe virtual channel state of corresponding basic packet switch router, VCS state store corresponding virtual circuit and exchange connection status, work as inputWhen for VCS signal, then the corresponding output virtual channel of input virtual channel is directly found by the information in the storage of VCS stateNumber, VCS signal is directly output to the corresponding PCIe device of output virtual channel number, and each input unit increases a bypassUnit is crossed over to allow packet switched data piece to be directly inputted to the crossbar switch in channel;
The path-calculating element exists for establishing the port ID- routed path or address: Port routed path for transmissionThe data slice of packet switch connection provides path arbitration result;
The virtual channel allocation unit, for the path arbitration result according to the path-calculating element or according to PS shapeState stores information, for input data piece distribution output virtual channel number;
The circuit configuration unit is matched for providing the crossbar switch prestored in the data slice that circuit switching connects for transmissionConfidence breath;
The crossbar switch allocation unit stores information for storing preset circuit configuration by circuit configuration, directlyCrossbar switch is configured, the data slice that circuit switching is connected directly enters the intersection by by-pass switch and opensIt closes and crosses over unit;
The crossbar switch exports data packet for the arbitration scheduling result according to the input unit across unitTo the corresponding PCIe device of output virtual channel number.
Preferably, the computing resource pool includes one of following or a variety of: CPU, GPU, FPGA, IPU.
Preferably, the memory resource pool includes one of following or a variety of: SSD, HDD.
On the other hand, the present invention provides a kind of hardware reconstruction methods, and this method is using hardware described in any of the above-described kindReconfiguration system, comprising:
Step 1: PCIe bus being scanned by management node, acquisition is connected to PCIe crossbar switch in PCIe multibus and pondOn PCIe device port information, wherein the management node be preset cpu node;
Step 2: the remote PCI e crossbar switch that PCIe crossbar switch shared interface connects in pond where management node obtainsPCIe device port information;
Step 3: CPU, GPU, FPGA, IPU, SSD, HDD for specifying PCIe bus to connect according to user requirements analysis etc. is hardThe type and quantity of part equipment;
Step 4: the type and quantity of the hardware device according to needed for step 3 judge local computing and memory resource poolWhether interior resource is enough, and step 5 is transferred to if enough, not enough, is transferred to step 7;
Step 5: management node configures the configuration space of the PCIe crossbar switch in resource pool, according to depth-firstSearching algorithm is that specified PCIe bus configuration distributes required processor, PCIe crossbar switch in configured pool, and is connected in pondPrincipal and subordinate's bus of device port, the information such as command register are managed everywhere in PCIe crossbar switch;
Step 6: passing through PCIe data needed for the packet format of PCIeDMA or the PCIe link layer protocol of the standard building present inventionPacket accesses the equipment for being connected to specified PCIe bus, terminates the configuration of PCIe hardware reconstruction;
Step 7: management node configures the configuration space of the PCIe crossbar switch in resource pool, total for specified PCIeLine distributes available calculating and storage resource;
Step 8: management node configures remote PCI e crossbar switch, carries out empty friendship according to remote PCI e crossbar switchChannel, packet switch channel, circuit-switched channels configuration are changed, two kinds of connecting interfaces of transparent bridge and non-transparent bridge are supported, needed for connectionAdjacent calculating and memory resource pool;
Step 9: management node is needed for specified PCIe bus assignment local resource pond according to Depth Priority AlgorithmProcessor, processor needed for selected adjacent resource pool, in configured pool/remote PCI e crossbar switch, and it is connected to PCIe intersectionPrincipal and subordinate's bus of device port, the information such as command register are managed everywhere in switch, and are transferred to step 6.
The present invention has the beneficial effect that:
The present invention GPU, FPGA, GPU, IPU, SSD etc. can be calculated by PCIe multibus and crossbar switch technology andStorage resource is decoupled from server and is come out, and is realized and the layers such as interconnected in the pond of Heterogeneous Computing and memory resource pool between interconnection, pondThe interconnection of secondaryization.The connection re-configurability of configuration offer standard PCIe computing resource based on PCIe crossbar switch is provided.Pass throughThe introducing of PCIe switch can create the high highly available system of cost performance, make full use of expanding by PCIe communicationMalleability, performance advantage, can support most of heterogeneous device, and versatility is stronger;PCIe configuration feature is provided, it can be according to data centerThe performance of server needs switching at runtime to calculate, storage and network, can be the calculate node of application distribution correct amount, each to countCalculating storage resource can be managed collectively, flexible allocation, support connection reconstruct;Data transfer delay, system complex can be reducedProperty and power consumption, Save power consumption rate is up to 50% in data-intensive environment.
The above description is only an overview of the technical scheme of the present invention, in order to better understand the technical means of the present invention,And it can be implemented in accordance with the contents of the specification, and in order to allow above and other objects of the present invention, feature and advantage canIt is clearer and more comprehensible, the followings are specific embodiments of the present invention.
Detailed description of the invention
By reading the following detailed description of the preferred embodiment, various other advantages and benefits are common for this fieldTechnical staff will become clear.The drawings are only for the purpose of illustrating a preferred embodiment, and is not considered as to the present inventionLimitation.And throughout the drawings, the same reference numbers will be used to refer to the same parts.In the accompanying drawings:
Fig. 1 is a kind of hardware reconstruction system architecture schematic diagram of the embodiment of the present invention;
Fig. 2 be the embodiment of the present invention it is a kind of based in pond/the specific framework of hardware reconstruction system of long-range crossbar switch showsIt is intended to;
Fig. 3 is the reconstruct schematic diagram based on PCIe crossbar switch in the resource pool of the embodiment of the present invention;
Fig. 4 is the reconstruct schematic diagram based on remote PCI e crossbar switch between the resource pool of the embodiment of the present invention;
Fig. 5 is the remote PCI e cross bar structure schematic diagram of the embodiment of the present invention;
Fig. 6 is the PCIe transaction packet package head format schematic diagram of the embodiment of the present invention;
Fig. 7 is the flow diagram of PCIe hardware reconstruction of embodiment of the present invention configuration.
Specific embodiment
Exemplary embodiments of the present disclosure are described in more detail below with reference to accompanying drawings.Although showing the disclosure in attached drawingExemplary embodiment, it being understood, however, that may be realized in various forms the disclosure without should be by embodiments set forth hereIt is limited.On the contrary, these embodiments are provided to facilitate a more thoroughly understanding of the present invention, and can be by the scope of the present disclosureIt is fully disclosed to those skilled in the art.
First embodiment of the invention provides a kind of hardware reconstruction system, referring to Fig. 1, comprising:
Between each resource pool and the resource pool closed on, the multichannel interface connection based on remote PCI e crossbar switch, resourceMultichannel interface connection in pond based on PCIe crossbar switch, remote PCI e crossbar switch and the PCIe crossbar switch in resource pool,It is connected together by shared interface, forms the double-deck crossbar configuration based on PCIe crossbar switch and remote PCI e crossbar switchLink;
Wherein, the resource pool includes computing resource pool and/or memory resource pool, and each processor passes through resource in resource poolPCIe crossbar switch inside pond is interconnected, and when being attached configuration, is configured inside resource pool by configuration informationPCIe crossbar switch connection relationship is calculating and storage resource inside the specified PCIe bus assignment resource pool of CPU, realizes meterIt calculates and is directly connected between resource pool, memory resource pool processor, support computing resource and the distribution of storage resource dynamic on-demand;
The remote PCI e crossbar switch is interconnected by shared interface and each resource pool, realizes the resource based on PCIe protocolPoint-to-point peer switch between pond configures the connection of remote PCI e crossbar switch by configuration information when being attached configurationRelationship is the processor resource in the specified PCIe bus assignment Adjacent resource pond of CPU, carries out calculating and storage money between resource poolSource dynamically distributes.
The present invention GPU, FPGA, GPU, IPU, SSD etc. can be calculated by PCIe multibus and crossbar switch technology andStorage resource is decoupled from server and is come out, and is realized and the layers such as interconnected in the pond of Heterogeneous Computing and memory resource pool between interconnection, pondThe interconnection of secondaryization.The connection re-configurability of configuration offer standard PCIe computing resource based on PCIe crossbar switch is provided.Pass throughThe introducing of PCIe switch makes full use of the scalability and performance advantage communicated by PCIe, can create cost performance heightHighly available system, can support most of heterogeneous device, versatility is stronger;PCIe configuration feature is provided, it can be according in dataThe performance of central server needs switching at runtime to calculate, storage and network, can be the calculate node of application distribution correct amount, respectivelyCalculating storage resource can be managed collectively, flexible allocation, support connection reconstruct;It is multiple that data transfer delay, system can be reducedPolygamy and power consumption, Save power consumption rate is up to 50% in data-intensive environment.
It should be noted that shared interface described in the embodiment of the present invention supports two kinds of interface moulds of transparent bridge and non-transparent bridgeFormula can be configured to master port or from port, interconnect between realizing port by PCIe shared interface cable.
In the embodiment of the present invention, the PCIe crossbar switch in the resource pool is with the realization of PCIe switching backplane, PCIe exchangeBottom plate includes multichannel PCIe bus slot and PCIe exchange chip.
In the embodiment of the present invention, the remote PCI e crossbar switch is realized with PCIe switch, realizes the point between resource poolTo an exchanged form, including one of following or a variety of: packet switch, circuit switching and virtual channel exchange.
The exchange of virtual channel described in the embodiment of the present invention multiple shares the same physics for being made up of virtual channelThe virtual circuit in channel exchanges connection, and virtual circuit exchange connection can connect cooperation with packet switch connection and circuit switching and comeTransmit flow data.
System described in the embodiment of the present invention further includes compiler;The compiler, for according to known internuclear circulationBelieve situation, be every known flow allocation of communications path, and determines that virtual circuit exchange connection, circuit switching connection and packet switch connectIt connects, is connected at runtime according to communication compiling as a result, establishing each exchange by the link information prestored, between different resource pondCommunication is transmitted in corresponding exchange connection.
In the embodiment of the present invention, the remote PCI e crossbar switch includes multiple input units, and a crossbar switch is crossed overUnit, path-calculating element, virtual channel allocation unit, crossbar switch allocation unit and circuit configuration unit;
The input unit, for the virtual channel number after the PCIe device configuration output arbitration for connection, each inputUnit includes n output virtual channel VC1-VCn, and a bypass channel, a PS state stores and a VCS state storage,The input unit exchanges VCS signal with the input virtual channel of PCIe device and port input signal is connected, the storage of PS stateThe virtual channel state of corresponding basic packet switch router, VCS state store corresponding virtual circuit and exchange connection status, work as inputWhen for VCS signal, then the corresponding output virtual channel of input virtual channel is directly found by the information in the storage of VCS stateNumber, VCS signal is directly output to the corresponding PCIe device of output virtual channel number, and each input unit increases a bypassUnit is crossed over to allow packet switched data piece to be directly inputted to the crossbar switch in channel;
The path-calculating element exists for establishing the port ID- routed path or address: Port routed path for transmissionThe data slice of packet switch connection provides path arbitration result;
The virtual channel allocation unit, for the path arbitration result according to the path-calculating element or according to PS shapeState stores information, for input data piece distribution output virtual channel number;
The circuit configuration unit is matched for providing the crossbar switch prestored in the data slice that circuit switching connects for transmissionConfidence breath;
The crossbar switch allocation unit stores information for storing preset circuit configuration by circuit configuration, directlyCrossbar switch is configured, the data slice that circuit switching is connected directly enters the intersection by by-pass switch and opensIt closes and crosses over unit;
The crossbar switch exports data packet for the arbitration scheduling result according to the input unit across unitTo the corresponding PCIe device of output virtual channel number.
It should be noted that computing resource pool described in the embodiment of the present invention includes one of following or a variety of: CPU,GPU, FPGA, IPU, etc..The memory resource pool includes one of following or a variety of: SSD, HDD, etc..Specific abilityThe technical staff in domain can be configured according to actual needs, and the present invention is not especially limited this.
Detailed explanation and illustration is carried out below in conjunction with Fig. 1-6 system described in the embodiment of the present invention:
As shown in Figure 1, the present invention constructs a series of resource pools first, and between each resource pool and the resource pool closed on, baseIt is connected in the multichannel interface of remote PCI e crossbar switch, the multichannel interface connection in resource pool based on PCIe crossbar switch, remotelyPCIe crossbar switch in PCIe crossbar switch and resource pool, is connected together by shared interface, formed based on crossbar switch andThe double-deck crossbar configuration link of remote PCI e crossbar switch supports star topology, network topology, tree-shaped between each resource poolA variety of connection relationships such as topology and full connection topology;
Wherein, resource pool includes computing resource pool and/or memory resource pool, and each processor passes through in resource pool in resource poolThe crossbar switch in portion is interconnected.Computing resource pool includes the Heterogeneous Computings resource such as CPU, GPU, FPGA, IPU, memory resource poolThe storage resources such as including SSD, HDD.Each calculating memory resource pool contains the pond CPU, the pond GPU, the pond FPGA, the pond IPU, storage poolDeng at least one resource pool.Wherein: the pond CPU contains muti-piece CPU processor, and the pond GPU contains muti-piece GPU processor, FPGAPond contains muti-piece FPGA processor, and the pond IPU contains the IPU processor such as muti-piece Cambrian, storage pool contain muti-piece SSD,The hard-disc storages resource such as HDD, each calculating and storage resource externally pass through PCIe bus and are attached with crossbar switch.ConnectedWhen connecing configuration, the crossbar switch connection relationship inside resource pool is configured by configuration information, is the specified PCIe bus assignment of CPUCalculating and storage resource inside resource pool are realized and are directly connected between computing resource pool, memory resource pool inner treater, are propped upHold computing resource and the distribution of storage resource dynamic on-demand.
Crossbar switch in resource pool includes multiple shared interfaces, and shared interface support transparent bridge and two kinds of non-transparent bridge connectMouth mold formula can be configured to master port or from port, interconnect, can connect between realizing port by PCIe shared interface cableThe PCIe device of arbitrary standards PCIe3.0 interface.Shared interface crossbar switch by multichannel PCIe bus and the pond CPU, the pond GPU,The pond FPGA, the pond IPU, in storage pool everywhere in reason device be directly connected to, configuration information can be passed through and control specified input interface and outputThe direct-connected relationship of interface can configure crossbar switch and specify PCIe bus assignment by the way that reconfiguration information is inputted crossbar switchIt calculates, storage resource, is calculated for CPU, FPGA, GPU, IPU, the memory etc. of multichannel PCIe bus assignment cascade on the busAnd storage resource.
Remote PCI e crossbar switch realizes point-to-point peer switch between the resource pool based on PCIe protocol.It is long-range to intersectSwitch is attached by shared interface and the crossbar switch in resource pool, realizes the communication interconnection between resource pool.ConnectedWhen connecing configuration, long-range crossbar switch connection relationship is configured by configuration information, is the specified PCIe bus assignment Adjacent resource of CPUThe processor resource in pond supports calculating and storage resource between resource pool to dynamically distribute.
By input reconstruct configuration information to crossbar switch, crossbar switch occupies feelings according to user demand and current systemCondition changes the connection relationship of the virtual PCI Bridge in crossbar switch by configuration information, is connected to it for specified PCIe bus assignmentIn calculating and storage resource.Such as a moment specifies the system 1 in PCIe bus for 4 pieces of GPU, 4 pieces of CPU, 4 pieces of FPGA, 8Block SSD;Subsequent time configures crossbar switch by reconfiguration information, and the computing system connected in the PCIe bus becomes system 2,Including 2 pieces of CPU, 1 piece of GPU, 1 piece of FPGA, 2 pieces of SSD;Subsequent time again configures crossbar switch by reconfiguration information, and the PCIe is totalThe computing system connected on line becomes system 3, including 1 piece of CPU, 1 piece of GPU, 1 piece of FPGA, 2 pieces of SSD.Pass through change to be formedThe connection relationship between storage resource is calculated, so that hardware system supports computing capability and concatenation ability reconstruct, to be formedIt calculates and the dynamic on-demand of storage resource distributes.
As shown in Fig. 2, the embodiment of the present invention includes 6 calculating and memory resource pool, 1 remote PCI e crossbar switch and 1A main controlled node.Wherein: including 1 piece of PCIe crossbar switch backboard in each resource pool, it includes multichannel PCIe that PCIe, which intersects backboard,Bus slot and 1 PEX8796 exchange chip, the calculating such as CPU, GPU, FPGA, IPU, SSD and storage resource insertion PCIe intersectIn the slot for switching backboard, the interconnection reconstruct in resource pool is realized.By PCIe shared interface connecting line and far between resource poolJourney crossbar switch is connected, and the MiniSAS interface cable of the shared interface connection line options PCIex4 of present case is realized.It is long-range to intersectSwitch realizes that the PCIe switch of the implementation case includes 12 ports in the form of PCIe switch, can support 12 metersIt calculates and the point-to-point peer switch of memory resource pool, 12 ports can be configured to transparent bridge mode and non-transparent bridge mode,Transparent bridge mode for connect calculating/memory resource pool without CPU, non-transparent bridge mode be used to connect include host calculating/Memory resource pool.Main controlled node is connected by remote PCI e crossbar switch with multiple calculating and memory resource pool, for reconstructingConfiguration and Data Transmission Controlling.Main controlled node is by PCIe crossbar switch in remote PCI e crossbar switch and resource pool to connectionHardware device is scanned, by PCIe crossbar switch in resource pool, remote PCI e crossbar switch according to reconstruct configuration informationIt is configured, distributes different calculating and storage resource for CPU, realize hardware system reconstruct.
As shown in figure 3, the reconfiguration system based on crossbar switch in the resource pool of another specific implementation case of the invention,Wherein, crossbar switch is realized in the form of PCIe switching backplane in resource pool, and PCIe switching backplane includes that multichannel PCIe bus is insertedSlot and PCIe exchange chip.It will be in crossbar switch backboard slot in different calculating and storage resource insertion pond.The resource poolHardware structure dynamic construction uses the method based on PCIe multiple bus and crossbar switch, and configuration information controls the defeated of crossbar switchThe interconnecting relation of incoming interface and output interface.The hardware resources such as difference CPU, GPU, FPGA, IPU, SSD, HDD are connected in figureThe different PCIe bus slot of crossbar switch, CPU, which can choose, to be connected to the hardware device group of specified bus slot and builds up differenceComputing system.By configuring the crossbar switch in resource pool, the hardware device connected in PCIe bus can change.ExampleSuch as: for subsequent time by changing crossbar configuration, the hardware resource of the PCIe buses such as PCIe bus 1, PCIe bus 2 connection canTo be exchanged, additions and deletions etc. change, to form new computing system.
For the hardware resources such as CPU, GPU, FPGA, IPU, SSD, HDD in different calculating and memory resource poolReconstruction Mechanism is as shown in Figure 4.It is interconnected between Adjacent resource pond by remote PCI e crossbar switch, is handled between realization resource poolThe communication of device.Remote PCI e crossbar switch is realized in the form of PCIe switch, including multiple PCIe shared ports, different resourceThe hardware structure dynamic construction in pond is realized based on the control of remote PCI e crossbar switch.
Such as: the hardware resource of resource pool 1 and resource pool 2 is carried out by remote PCI e crossbar switch shown in Fig. 4 in figureConnection allows resource pool 1 and the hardware resource of resource 2 directly can realize data interaction by the transmission of PCIe.Remote PCI e intersectsSwitch is connected together with the crossbar switch in resource pool by shared interface.It is formed and is opened based on crossbar switch and remote PCI e intersectionThe double-deck crossbar configuration link of pass.When being attached configuration, calculating and storage resource are configured by configuration information firstCrossbar switch connection relationship inside pond calculates with the specified PCIe bus assignment inside memory resource pool and stores money to calculateSource;Then the remote PCI e crossbar switch between resource pool is configured, realizes that resource pool is directly connected to resource pool.As shown in Figure 4Resource pool between hardware reconstruction mechanism, can be made by the configuration to crossbar switch and remote PCI e crossbar switch in pondObtain may be implemented to be in communication with each other between SSD, CPU in CPU, FPGA and mainboard 2 in mainboard 1, without influencing other hardware moneyThe use in source.
Fig. 5 is the composed structure schematic diagram of long-range crossbar switch, as shown in figure 5, the embodiment of the present invention is supported in followingOne or more exchanged forms: packet switch PS, circuit switching (CS) and virtual channel exchange VCS.Virtual channel exchange, for passing throughVirtual channel forms multiple virtual circuits exchange connections for sharing the same physical channel, and virtual circuit exchange connection can be withCooperation is connect with packet switch connection and circuit switching to transmit flow data.According to known internuclear stream communication situation, pass through compilingDevice is every known flow allocation of communications path, and determines that virtual circuit exchange connection, circuit switching connection are connected with packet switch,It is established respectively according to communication compiling as a result, link information is stored in each PCIe exchanger by the link information prestored when operationExchange connects, and the communication between different motherboard resources ponds is transmitted in corresponding exchange connection.
Compared with basic packet switcher, the exchanger of mixed strategy of the present invention needs some additional hardware: including sidePaths, circuit configuration storage and the storage of VCS state.Remote PCI e crossbar switch includes multiple input units, and an intersection is openedIt closes and crosses over, path computing, the submodules such as virtual channel distribution, crossbar switch distribution, circuit configuration.
Wherein: input unit is used for the virtual channel number after the PCIe device configuration output arbitration for connection, each inputUnit includes n output virtual channel VC1-VCn, a bypass channel, a PS state storage and a VCS state storage.The input unit exchanges VCS signal and port input signal with the input virtual channel of PCIe device and is connected, VCS signal and defeatedThe port numbers of the number and PCIe device that enter signal correspond.Virtual the leading to of the corresponding basic packet switch router of PS state storageRoad state.VCS state stores corresponding virtual circuit and exchanges connection status.When input is VCS signal, then directly pass through VCS shapeInformation in state storage finds the corresponding output virtual channel number of input virtual channel, and it is empty that VCS signal is directly output to outputThe quasi- corresponding PCIe device of channel number;PS state and VCS state have all respectively contained n domain, this corresponding n virtual channel.AndAnd n virtual channel is connected with packet switch by virtual circuit exchange connection and is shared.And each input unit increases a bypassChannel allows data slice not need through virtual channel, is directly inputted to crossbar switch across submodule.
Path calculation module is attached with input unit, for establishing the port ID- routed path or address: Port routingPath provides path arbitration result in the data slice that packet switch connects for transmission.Wherein ID- port routing table provides PCIe deviceMap paths relationship number between port numbers;Address: Port routing table is provided between PCIe device physical address and port numbersMap paths relationship.Communication path arbitration is carried out according to the ID in PCIe transaction packet, address when operation.
Virtual channel allocation unit stores letter for the path arbitration result according to path-calculating element or according to PS stateBreath, for input data piece distribution output virtual channel number;
Circuit configuration unit, for providing the crossbar configuration prestored letter in the data slice that circuit switching connects to transmitBreath, storage are used to connect the link information of storage physical channel to circuit switching, distribute with crossbar switch distribution and virtual channelModule is attached, and circuit switching connection is established by setting circuit configuration information.
Crossbar switch allocation unit stores information for storing preset circuit configuration by circuit configuration, directly to friendshipFork puts capable configuration into, and the data slice that circuit switching is connected directly enters crossbar switch across son by by-pass switchModule;
Data packet is output to output for the arbitration scheduling result according to input unit across submodule by crossbar switchThe corresponding PCIe device of virtual channel number.
For the mixed strategy for realizing support circuit-switched, packet switch and virtual channel exchange, this patent has modified PCIe thingIt is engaged in the PCIe transaction layer packet header wrapped, exchange flag bit being added in reserved field, as shown in Figure 6.Carrying out PCIe crossbar switchWhen hardware reconstruction configures, according to the exchange flag bit in PCIe transaction layer packet header, corresponding exchanging policy is selected.
If the exchange flag bit in input signal PCIe transaction layer packet header is input virtual channel VCS signal, directly logicalVCS status information is crossed, finds the corresponding output virtual channel number of input virtual channel, it is empty that VCS signal is directly output to outputThe quasi- corresponding PCIe device of channel number.
If input signal is PCIe port input signal, needs to be inputted in input unit according to PCIe port first and believeNumber carry out arbitration scheduling.It is that transmission is handed in circuit if the exchange flag bit in PCIe transaction layer packet header is circuit-switched signalsThe data slice for changing connection reaches input unit, and crossbar switch is stood across submodule according to the link information prestored in circuit configurationIt configures, then the data slice directly passes through bypass channel and enters crossbar switch across submodule, and crossbar switch crosses over submoduleData packet is directly output to corresponding PCIe device according to configuration result.
If input signal is PCIe port input signal, needs to be inputted in input unit according to PCIe port first and believeNumber carry out arbitration scheduling.It is that transmission connects in packet switch if the exchange flag bit in PCIe transaction layer packet header is packet switch signalThe data slice that connects reaches input unit, then first according to the path arbitration result of path-calculating element or according to PS status information,Output virtual channel number is distributed for input data piece, output virtual port number is then sent to crossbar switch and crosses over submodule,Data packet, according to arbitration scheduling result, is output to the corresponding PCIe of output virtual channel number and set by crossbar switch across submoduleIt is standby.
Second embodiment of the invention provides a kind of hardware reconstruction method, and referring to Fig. 7, this method application present invention first is realApply any hardware reconstruction system in example, comprising:
Step 1: PCIe bus being scanned by management node, acquisition is connected to PCIe crossbar switch in PCIe multibus and pondOn PCIe device port information, wherein the management node be preset cpu node;
Step 2: the remote PCI e crossbar switch that PCIe crossbar switch shared interface connects in pond where management node obtainsPCIe device port information;
Step 3: CPU, GPU, FPGA, IPU, SSD, HDD for specifying PCIe bus to connect according to user requirements analysis etc. is hardThe type and quantity of part equipment;
Step 4: the type and quantity of the hardware device according to needed for step 3 judge local computing and memory resource poolWhether interior resource is enough, and step 5 is transferred to if enough, not enough, is transferred to step 7;
Step 5: management node configures the configuration space of the PCIe crossbar switch in resource pool, according to depth-firstSearching algorithm is that specified PCIe bus configuration distributes required processor, PCIe crossbar switch in configured pool, and is connected in pondPrincipal and subordinate's bus of device port, the information such as command register are managed everywhere in PCIe crossbar switch;
Step 6: passing through PCIe data needed for the packet format of PCIeDMA or the PCIe link layer protocol of the standard building present inventionPacket accesses the equipment for being connected to specified PCIe bus, terminates the configuration of PCIe hardware reconstruction;
Step 7: management node configures the configuration space of the PCIe crossbar switch in resource pool, total for specified PCIeLine distributes available calculating and storage resource;
Step 8: management node configures remote PCI e crossbar switch, carries out empty friendship according to remote PCI e crossbar switchChannel, packet switch channel, circuit-switched channels configuration are changed, two kinds of connecting interfaces of transparent bridge and non-transparent bridge are supported, needed for connectionAdjacent calculating and memory resource pool;
Step 9: management node is needed for specified PCIe bus assignment local resource pond according to Depth Priority AlgorithmProcessor, processor needed for selected adjacent resource pool, in configured pool/remote PCI e crossbar switch, and it is connected to PCIe intersectionPrincipal and subordinate's bus of device port, the information such as command register are managed everywhere in switch, and are transferred to step 6.
The embodiment of the present invention is realized by the way that configuration is reconstructed to crossbar switch, remote PCI e crossbar switch in resource poolComputing capability reconstruct, concatenation ability reconstruct.For details, reference can be made to first embodiment of the invention, the embodiment of the present invention is not made this in detailIt discusses.
The embodiment of the present invention can at least bring it is below the utility model has the advantages that
1. the embodiment of the present invention supports the Heterogeneous Computings resources such as CPU, GPU, FPGA, IPU, the storage resources such as SSD, HDD pointCloth is interconnected between each memory resource pool by remote PCI e crossbar switch, storage resource in each calculating and memory resource poolEach CPU, GPU in pond, FPGA, IPU, it is interconnected by PCIe multibus and crossbar switch between storage resource, so that hardware resourceCan be decoupling, improve the dynamic configurability of network;
2. the present invention provides the stratification interconnection scheme of PCIe crossbar switch and remote PCI e crossbar switch in a kind of pond, lead toIt crosses and is configured in pond with remote PCI e crossbar switch, can be needed according to the performance of data center server for application distribution justWhen the calculate node and memory node of quantity, each storage resource that calculates can be managed collectively, and flexible allocation is each hard by reconstructingConnection between part resource can effectively promote hardware frame so that hardware structure supports computing capability and concatenation ability to construct on demandThe performance and efficiency of structure;
3. the present invention provides a kind of remote PCI e crossbar configuration scheme of hybrid switching mechanism, while packet being supported to hand overIt changes, circuit switching and virtual channel three kinds of exchanged forms of exchange, wherein virtual channel exchange can directly pass through crossbar switch, configureIt is more easy, while supporting to use packet switch or circuit switching in link conflict, guarantee efficiently to be based on PCIe multiple busWith the communication link of crossbar switch.
It should be noted that PCIe crossbar switch described in the embodiment of the present invention can be realized by PCIeswitch chip,Crossbar switch function can also be realized by FPGA.
Algorithm and display are not inherently related to any particular computer, virtual system, or other device provided herein.Various general-purpose systems can also be used together with teachings based herein.As described above, it constructs required by this kind of systemStructure be obvious.In addition, the present invention is also not directed to any particular programming language.It should be understood that can use variousProgramming language realizes summary of the invention described herein, and the description done above to language-specific is to disclose this hairBright preferred forms.
In the instructions provided here, numerous specific details are set forth.It is to be appreciated, however, that implementation of the inventionExample can be practiced without these specific details.In some instances, well known method, structure is not been shown in detailAnd technology, so as not to obscure the understanding of this specification.
Similarly, it should be understood that in order to simplify the disclosure and help to understand one or more of the various inventive aspects,Above in the description of exemplary embodiment of the present invention, each feature of the invention is grouped together into single implementation sometimesIn example, figure or descriptions thereof.However, the disclosed method should not be interpreted as reflecting the following intention: i.e. required to protectShield the present invention claims features more more than feature expressly recited in each claim.More precisely, as followingClaims reflect as, inventive aspect is all features less than single embodiment disclosed above.Therefore,Thus the claims for following specific embodiment are expressly incorporated in the specific embodiment, wherein each claim itselfAll as a separate embodiment of the present invention.
Those skilled in the art will understand that can be carried out adaptively to the module in the equipment in embodimentChange and they are arranged in one or more devices different from this embodiment.It can be the module or list in embodimentMember or component are combined into a module or unit or component, and furthermore they can be divided into multiple submodule or subelement orSub-component.Other than such feature and/or at least some of process or unit exclude each other, it can use anyCombination is to all features disclosed in this specification (including adjoint claim, abstract and attached drawing) and so disclosedAll process or units of what method or apparatus are combined.Unless expressly stated otherwise, this specification is (including adjoint powerBenefit require, abstract and attached drawing) disclosed in each feature can carry out generation with an alternative feature that provides the same, equivalent, or similar purposeIt replaces.
In addition, it will be appreciated by those of skill in the art that although some embodiments described herein include other embodimentsIn included certain features rather than other feature, but the combination of the feature of different embodiments mean it is of the inventionWithin the scope of and form different embodiments.For example, in the following claims, embodiment claimed is appointedMeaning one of can in any combination mode come using.
Various component embodiments of the invention can be implemented in hardware, or to run on one or more processorsSoftware module realize, or be implemented in a combination thereof.It will be understood by those of skill in the art that can be used in practiceMicroprocessor or digital signal processor (DSP) realize that Distributed File System Data according to an embodiment of the present invention importsThe some or all functions of some or all components in device.The present invention is also implemented as being retouched here for executingThe some or all device or device programs (for example, computer program and computer program product) for the method stated.It is such to realize that program of the invention can store on a computer-readable medium, or can have one or more signalForm.Such signal can be downloaded from an internet website to obtain, be perhaps provided on the carrier signal or with it is any itsHe provides form.
It should be noted that the above-mentioned embodiments illustrate rather than limit the invention, and abilityField technique personnel can be designed alternative embodiment without departing from the scope of the appended claims.In the claims,Any reference symbol between parentheses should not be configured to limitations on claims.Word "comprising" does not exclude the presence of notElement or step listed in the claims.Word "a" or "an" located in front of the element does not exclude the presence of multiple suchElement.The present invention can be by means of including the hardware of several different elements and being come by means of properly programmed computer realIt is existing.In the unit claims listing several devices, several in these devices can be through the same hardware branchTo embody.The use of word first, second, and third does not indicate any sequence.These words can be explained and be run after fameClaim.

Claims (10)

Translated fromChinese
1.一种硬件重构系统,其特征在于,包括:1. a hardware reconstruction system, is characterized in that, comprises:每个资源池与临近的资源池之间,基于远程PCIe交叉开关的多路接口连接,资源池内基于PCIe交叉开关的多路接口连接,远程PCIe交叉开关与资源池内的PCIe交叉开关,通过共享接口连在一起,形成基于PCIe交叉开关和远程PCIe交叉开关的双层交叉开关配置链路;Between each resource pool and the adjacent resource pool, the multi-channel interface connection based on the remote PCIe crossbar is connected, and the multi-channel interface connection based on the PCIe crossbar in the resource pool, the remote PCIe crossbar and the PCIe crossbar in the resource pool are connected through the shared interface. Together, a dual-layer crossbar configuration link based on PCIe crossbar and remote PCIe crossbar is formed;其中,所述资源池包括计算资源池和/或存储资源池,资源池内各处理器通过资源池内部的PCIe交叉开关进行互联,在进行连接配置时,通过配置信息配置资源池内部的PCIe交叉开关连接关系,为CPU的指定PCIe总线分配资源池内部的计算和存储资源,实现计算资源池、存储资源池内处理器之间的直接连接,支持计算资源和存储资源动态按需分配;The resource pool includes a computing resource pool and/or a storage resource pool, each processor in the resource pool is interconnected through a PCIe crossbar inside the resource pool, and during connection configuration, the PCIe crossbar inside the resource pool is configured through configuration information Connection relationship: Allocate computing and storage resources within the resource pool for the specified PCIe bus of the CPU, realize direct connection between processors in the computing resource pool and storage resource pool, and support dynamic on-demand allocation of computing resources and storage resources;所述远程PCIe交叉开关通过共享接口与各资源池互联,实现基于PCIe协议的资源池之间点对点的对等交换,在进行连接配置时,通过配置信息配置远程PCIe交叉开关连接关系,为CPU的指定PCIe总线分配相邻资源池的处理器资源,进行资源池间的计算和存储资源动态分配。The remote PCIe cross switch is interconnected with each resource pool through a shared interface to realize point-to-point peer-to-peer exchange between resource pools based on the PCIe protocol. During connection configuration, the connection relationship of the remote PCIe cross switch is configured through the configuration information, which is designated by the CPU. The PCIe bus allocates processor resources in adjacent resource pools, and dynamically allocates computing and storage resources between resource pools.2.根据权利要求1所述的系统,其特征在于,2. The system of claim 1, wherein:所述共享接口支持透明桥和非透明桥两种接口模式,均可被配置为主端口或从端口,通过PCIe共享接口线缆实现端口间互联。The shared interface supports two interface modes, a transparent bridge and a non-transparent bridge, and can be configured as a master port or a slave port, and the ports are interconnected through a PCIe shared interface cable.3.根据权利要求1所述的系统,其特征在于,3. The system of claim 1, wherein:所述资源池内的PCIe交叉开关以PCIe交换底板实现,PCIe交换底板包括多路PCIe总线插槽和PCIe交换芯片。The PCIe cross switch in the resource pool is implemented by a PCIe switch backplane, and the PCIe switch backplane includes multiple PCIe bus slots and PCIe switch chips.4.根据权利要求1所述的系统,其特征在于,4. The system of claim 1, wherein:所述远程PCIe交叉开关以PCIe交换机实现,实现资源池间的点对点交换方式,包括以下中的一种或多种:包交换、电路交换和虚拟通道交换。The remote PCIe cross switch is implemented by a PCIe switch, and implements a point-to-point switching mode between resource pools, including one or more of the following: packet switching, circuit switching, and virtual channel switching.5.根据权利要求4所述的系统,其特征在于,5. The system of claim 4, wherein:所述虚拟通道交换,用于通过虚拟通道组成多个可共享同一个物理通道的虚拟电路交换连接,且虚拟电路交换连接可以与包交换连接和电路交换连接配合来传输流数据。The virtual channel switching is used to form a plurality of virtual circuit switching connections that can share the same physical channel through the virtual channel, and the virtual circuit switching connection can cooperate with the packet switching connection and the circuit switching connection to transmit stream data.6.根据权利要求4所述的系统,其特征在于,还包括编译器;6. The system of claim 4, further comprising a compiler;所述编译器,用于根据已知的核间流通信情况,为每条已知流通信分配路径,并确定虚拟电路交换连接、电路交换连接和包交换连接,在运行时根据通信编译结果,通过预存的连接信息建立各交换连接,不同资源池之间的通信在各自对应的交换连接上进行传输。The compiler is configured to allocate a path for each known flow communication according to the known inter-core flow communication situation, and determine the virtual circuit-switched connection, the circuit-switched connection and the packet-switched connection, and compile the result according to the communication at runtime, Each switching connection is established through the pre-stored connection information, and the communication between different resource pools is transmitted on the corresponding switching connection.7.根据权利要求1所述的系统,其特征在于,7. The system of claim 1, wherein:所述远程PCIe交叉开关包括多个输入单元,一个交叉开关跨越单元,路径计算单元,虚拟通道分配单元、交叉开关分配单元和电路配置单元;The remote PCIe crossbar includes a plurality of input units, a crossbar crossover unit, a path calculation unit, a virtual channel assignment unit, a crossbar assignment unit and a circuit configuration unit;所述输入单元,用于为连接的PCIe设备配置输出仲裁后的虚拟通道号,每个输入单元包括n个输出虚拟通道VC1-VCn,一个旁路通道,一个PS状态存储和一个VCS状态存储,所述输入单元与PCIe设备的输入虚拟通道交换VCS信号和端口输入信号相连,PS状态存储对应基本包交换路由器的虚拟通道状态,VCS状态存储对应虚拟电路交换连接状态,当输入为VCS信号时,则直接通过VCS状态存储中的信息,找到输入虚拟通道对应的输出虚拟通道号,VCS信号直接被输出到输出虚拟通道号对应的PCIe设备,且每个输入单元增加一个旁路通道来允许包交换数据片直接输入到所述交叉开关跨越单元;The input unit is used to configure the virtual channel number after output arbitration for the connected PCIe device, and each input unit includes n output virtual channels VC1-VCn, a bypass channel, a PS state storage and a VCS state storage, The input unit is connected with the input virtual channel switching VCS signal of the PCIe device and the port input signal, the PS state stores the virtual channel state of the corresponding basic packet switching router, and the VCS state stores the corresponding virtual circuit switching connection state. When the input is a VCS signal, Then directly through the information in the VCS state storage, find the output virtual channel number corresponding to the input virtual channel, the VCS signal is directly output to the PCIe device corresponding to the output virtual channel number, and each input unit adds a bypass channel to allow packet switching Data slices are directly input to the crossbar spanning unit;所述路径计算单元,用于建立ID-端口路由路径或地址-端口路由路径,为传输在包交换连接的数据片提供路径仲裁结果;The path calculation unit is used to establish an ID-port routing path or an address-port routing path, and provide a path arbitration result for transmitting data slices connected in packet switching;所述虚拟通道分配单元,用于根据所述路径计算单元的路径仲裁结果或根据PS状态存储信息,为输入数据片分配输出虚拟通道号;The virtual channel assignment unit is configured to assign an output virtual channel number to the input data slice according to the path arbitration result of the path calculation unit or according to the PS state storage information;所述电路配置单元,用于为传输在电路交换连接的数据片提供预存的交叉开关配置信息;the circuit configuration unit, configured to provide pre-stored crossbar configuration information for data pieces transmitted in the circuit-switched connection;所述交叉开关分配单元,用于通过电路配置存储预设的电路配置存储信息,直接对交叉开关进行配置,使得电路交换连接的数据片可以直接通过旁路开关进入所述交叉开关跨越单元;The crossbar distribution unit is configured to store preset circuit configuration storage information through the circuit configuration, and directly configure the crossbar, so that the data slices connected by circuit switching can directly enter the crossbar crossover unit through the bypass switch;所述交叉开关跨越单元,用于根据所述输入单元的仲裁调度结果,将数据包输出到输出虚拟通道号对应的PCIe设备。The crossbar spanning unit is configured to output the data packet to the PCIe device corresponding to the output virtual channel number according to the arbitration scheduling result of the input unit.8.根据权利要求1所述的系统,其特征在于,8. The system of claim 1, wherein:所述计算资源池包括以下中的一种或多种:CPU、GPU、FPGA、IPU。The computing resource pool includes one or more of the following: CPU, GPU, FPGA, and IPU.9.根据权利要求1所述的系统,其特征在于,9. The system of claim 1, wherein:所述存储资源池包括以下中的一种或多种:SSD、HDD。The storage resource pool includes one or more of the following: SSD and HDD.10.一种硬件重构方法,其特征在于,应用权利要求1-9中任一项所述的硬件重构系统,包括:10. A hardware reconfiguration method, characterized in that, applying the hardware reconfiguration system according to any one of claims 1-9, comprising:步骤1:通过管理节点扫描PCIe总线,获取连接在PCIe多总线和池内PCIe交叉开关上的PCIe设备端口信息,其中,所述管理节点为预设的CPU节点;Step 1: Scan the PCIe bus by the management node, and obtain the PCIe device port information connected to the PCIe multi-bus and the PCIe crossbar in the pool, wherein the management node is a preset CPU node;步骤2:管理节点获取所在池内PCIe交叉开关共享接口连接的远程PCIe交叉开关的PCIe设备端口信息;Step 2: the management node obtains the PCIe device port information of the remote PCIe crossbar connected to the shared interface of the PCIe crossbar in the pool where it is located;步骤3:根据用户需求分析指定PCIe总线连接的CPU、GPU、FPGA、IPU、SSD、HDD等硬件设备的种类和数量;Step 3: Analyze and specify the type and quantity of hardware devices such as CPU, GPU, FPGA, IPU, SSD, and HDD connected to the PCIe bus according to user requirements;步骤4:根据步骤3中所需的硬件设备的种类和数量,判断本地计算和存储资源池内的资源是否够用,如果够用则转入步骤5,不够用则转入步骤7;Step 4: According to the type and quantity of hardware devices required in Step 3, determine whether the resources in the local computing and storage resource pool are sufficient, if sufficient, go to Step 5, and if not, go to Step 7;步骤5:管理节点对资源池内的PCIe交叉开关的配置空间进行配置,按照深度优先搜索算法为指定PCIe总线配置分配所需的处理器,配置池内PCIe交叉开关,和连接在池内PCIe交叉开关的各处理器端口的主从总线,命令寄存器等信息;Step 5: The management node configures the configuration space of the PCIe crossbars in the resource pool, allocates the required processors for the specified PCIe bus configuration according to the depth-first search algorithm, configures the PCIe crossbars in the pool, and connects to the PCIe crossbars in the pool. The master-slave bus, command register and other information of the processor port;步骤6:通过PCIe DMA或标准的PCIe链路层协议的包格式构建本发明所需PCIe数据包,访问连接在指定PCIe总线的设备,结束PCIe硬件重构配置;Step 6: construct the PCIe data packet required by the present invention through the packet format of the PCIe DMA or the standard PCIe link layer protocol, access the device connected to the specified PCIe bus, and end the PCIe hardware reconstruction configuration;步骤7:管理节点对资源池内的PCIe交叉开关的配置空间进行配置,为指定PCIe总线分配可用的计算和存储资源;Step 7: The management node configures the configuration space of the PCIe crossbar in the resource pool, and allocates available computing and storage resources for the designated PCIe bus;步骤8:管理节点对远程PCIe交叉开关进行配置,按照远程PCIe交叉开关进行虚交换通道、包交换通道、电路交换通道配置,支持透明桥和非透明桥两种连接接口,连接所需相邻的计算和存储资源池;Step 8: The management node configures the remote PCIe cross switch, configures virtual switching channels, packet switching channels, and circuit switching channels according to the remote PCIe cross switch, supports two connection interfaces of transparent bridge and non-transparent bridge, and connects the required adjacent computing and storage resource pools;步骤9:管理节点按照深度优先搜索算法为指定PCIe总线分配本地资源池所需的处理器,所选相邻资源池所需的处理器,配置池内/远程PCIe交叉开关,和连接在PCIe交叉开关的各处理器端口的主从总线,命令寄存器等信息,并转入步骤6。Step 9: The management node allocates the processors required by the local resource pool for the specified PCIe bus according to the depth-first search algorithm, the processors required by the selected adjacent resource pools, configures the in-pool/remote PCIe crossbar, and connects to the PCIe crossbar The master-slave bus, command register and other information of each processor port, and go to step 6.
CN201811113981.5A2018-09-252018-09-25Hardware reconfiguration system and methodActiveCN109240832B (en)

Priority Applications (1)

Application NumberPriority DateFiling DateTitle
CN201811113981.5ACN109240832B (en)2018-09-252018-09-25Hardware reconfiguration system and method

Applications Claiming Priority (1)

Application NumberPriority DateFiling DateTitle
CN201811113981.5ACN109240832B (en)2018-09-252018-09-25Hardware reconfiguration system and method

Publications (2)

Publication NumberPublication Date
CN109240832Atrue CN109240832A (en)2019-01-18
CN109240832B CN109240832B (en)2022-04-19

Family

ID=65056108

Family Applications (1)

Application NumberTitlePriority DateFiling Date
CN201811113981.5AActiveCN109240832B (en)2018-09-252018-09-25Hardware reconfiguration system and method

Country Status (1)

CountryLink
CN (1)CN109240832B (en)

Cited By (12)

* Cited by examiner, † Cited by third party
Publication numberPriority datePublication dateAssigneeTitle
CN111552562A (en)*2020-04-132020-08-18中国电子科技集团公司电子科学研究院Reconstruction configuration method and device for heterogeneous server
CN111880911A (en)*2020-06-192020-11-03浪潮电子信息产业股份有限公司Task load scheduling method, device and equipment and readable storage medium
CN112731823A (en)*2019-10-282021-04-30深圳市国微电子有限公司FPGA interconnection line circuit and FPGA interconnection line delay reduction method
CN113395216A (en)*2020-03-112021-09-14辉达公司Techniques to transfer data between hardware devices
CN113452731A (en)*2020-03-252021-09-28阿里巴巴集团控股有限公司Resource allocation method, device, electronic equipment and computer readable storage medium
CN114020452A (en)*2021-10-222022-02-08浪潮电子信息产业股份有限公司 A heterogeneous resource pool management method, system and heterogeneous computing platform
CN114445260A (en)*2022-01-172022-05-06苏州浪潮智能科技有限公司Distributed GPU communication method and device based on FPGA
WO2023098295A1 (en)*2021-11-302023-06-08中兴通讯股份有限公司Radio frequency chip, algorithm reconstruction method, and computer readable storage medium
CN116594785A (en)*2023-07-182023-08-15四川华鲲振宇智能科技有限责任公司Hardware-based server paravirtualization method
CN117687956A (en)*2024-01-312024-03-12苏州元脑智能科技有限公司 Multi-accelerator card heterogeneous server and resource link reconstruction method
WO2024088263A1 (en)*2022-10-252024-05-02杭州阿里云飞天信息技术有限公司Heterogeneous server system and use method thereof
CN119052199A (en)*2024-09-062024-11-29浪潮(北京)电子信息产业有限公司Equipment interconnection system, method, equipment, medium and program product

Citations (6)

* Cited by examiner, † Cited by third party
Publication numberPriority datePublication dateAssigneeTitle
US5850395A (en)*1995-07-191998-12-15Fujitsu Network Communications, Inc.Asynchronous transfer mode based service consolidation switch
US7181578B1 (en)*2002-09-122007-02-20Copan Systems, Inc.Method and apparatus for efficient scalable storage management
US20080229326A1 (en)*2007-01-262008-09-18Objective Interface Systems, Inc.Hardware communications infrastructure supporting location transparency and dynamic partial reconfiguration
CN101630305A (en)*2008-07-162010-01-20中国人民解放军信息工程大学Flexible management method for reconfigurable components in high-efficiency computer
CN106897581A (en)*2017-01-252017-06-27人和未来生物科技(长沙)有限公司A kind of restructural heterogeneous platform understood towards gene data
CN107786198A (en)*2016-08-252018-03-09富士施乐株式会社Reconfigurable logic circuit

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication numberPriority datePublication dateAssigneeTitle
US5850395A (en)*1995-07-191998-12-15Fujitsu Network Communications, Inc.Asynchronous transfer mode based service consolidation switch
US7181578B1 (en)*2002-09-122007-02-20Copan Systems, Inc.Method and apparatus for efficient scalable storage management
US20080229326A1 (en)*2007-01-262008-09-18Objective Interface Systems, Inc.Hardware communications infrastructure supporting location transparency and dynamic partial reconfiguration
CN101711467A (en)*2007-01-262010-05-19目标接口系统公司A hardware communications infrastructure supporting location transparency and dynamic partial reconfiguration
CN101630305A (en)*2008-07-162010-01-20中国人民解放军信息工程大学Flexible management method for reconfigurable components in high-efficiency computer
CN107786198A (en)*2016-08-252018-03-09富士施乐株式会社Reconfigurable logic circuit
CN106897581A (en)*2017-01-252017-06-27人和未来生物科技(长沙)有限公司A kind of restructural heterogeneous platform understood towards gene data

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
吴雄洲: "基于多层树形结构的PCIE总线配置技术研究", 《 航空计算技术》*
无: "Product Brief PEX8796. PLX Technology", 《HTTPS://HTML.ALLDATASHEET.COM/HTML-PDF/1079816/PLX/PEX8796/62/1/PEX8796.HTML》*

Cited By (20)

* Cited by examiner, † Cited by third party
Publication numberPriority datePublication dateAssigneeTitle
CN112731823A (en)*2019-10-282021-04-30深圳市国微电子有限公司FPGA interconnection line circuit and FPGA interconnection line delay reduction method
CN113395216B (en)*2020-03-112024-04-09辉达公司 Technology for transferring data between hardware devices
US11995023B2 (en)2020-03-112024-05-28NVIDIA Technologies, Inc.Techniques to transfer data among hardware devices
CN113395216A (en)*2020-03-112021-09-14辉达公司Techniques to transfer data between hardware devices
CN113452731B (en)*2020-03-252022-04-29阿里巴巴集团控股有限公司Resource allocation method, device, electronic equipment and computer readable storage medium
CN113452731A (en)*2020-03-252021-09-28阿里巴巴集团控股有限公司Resource allocation method, device, electronic equipment and computer readable storage medium
CN111552562B (en)*2020-04-132022-10-28中国电子科技集团公司电子科学研究院 Reconfiguration configuration method and device for heterogeneous servers
CN111552562A (en)*2020-04-132020-08-18中国电子科技集团公司电子科学研究院Reconstruction configuration method and device for heterogeneous server
CN111880911A (en)*2020-06-192020-11-03浪潮电子信息产业股份有限公司Task load scheduling method, device and equipment and readable storage medium
CN114020452A (en)*2021-10-222022-02-08浪潮电子信息产业股份有限公司 A heterogeneous resource pool management method, system and heterogeneous computing platform
WO2023098295A1 (en)*2021-11-302023-06-08中兴通讯股份有限公司Radio frequency chip, algorithm reconstruction method, and computer readable storage medium
CN114445260A (en)*2022-01-172022-05-06苏州浪潮智能科技有限公司Distributed GPU communication method and device based on FPGA
CN114445260B (en)*2022-01-172024-01-12苏州浪潮智能科技有限公司 FPGA-based distributed GPU communication method and device
WO2024088263A1 (en)*2022-10-252024-05-02杭州阿里云飞天信息技术有限公司Heterogeneous server system and use method thereof
CN116594785B (en)*2023-07-182023-09-15四川华鲲振宇智能科技有限责任公司Hardware-based server paravirtualization method
CN116594785A (en)*2023-07-182023-08-15四川华鲲振宇智能科技有限责任公司Hardware-based server paravirtualization method
CN117687956A (en)*2024-01-312024-03-12苏州元脑智能科技有限公司 Multi-accelerator card heterogeneous server and resource link reconstruction method
CN117687956B (en)*2024-01-312024-05-07苏州元脑智能科技有限公司 Multi-accelerator card heterogeneous server and resource link reconstruction method
CN119052199A (en)*2024-09-062024-11-29浪潮(北京)电子信息产业有限公司Equipment interconnection system, method, equipment, medium and program product
CN119052199B (en)*2024-09-062025-10-03浪潮(北京)电子信息产业有限公司 Device interconnection system, method, device, medium and program product

Also Published As

Publication numberPublication date
CN109240832B (en)2022-04-19

Similar Documents

PublicationPublication DateTitle
CN109240832A (en)A kind of hardware reconstruction system and method
JP6653366B2 (en) Computer cluster configuration for processing computation tasks and method for operating it
CN103117929B (en)A kind of communication means of Based PC Ie exchanges data and system
CN116501681A (en)CXL data transmission board card and method for controlling data transmission
CN104247341B (en)The dynamic optimization method of distribution switch and its multicast tree hierarchical structure
CN104871145A (en)Memory sharing in network device
CN106575283B (en) Cluster server configuration using cellular automata
CN105706404A (en) Method and apparatus for managing direct interconnect switch cabling and growth of a computer network
CN113986813B (en) Method, system, device and storage medium for construction and use of on-chip network structure
US12411795B2 (en)Enabling a multi-chip daisy chain topology using peripheral component interconnect express (PCIe)
GB2502861A (en)Network-on-chip comprising distributed reorder buffers
US10896476B2 (en)Repository of integration description of hardware intellectual property for NoC construction and SoC integration
US8352695B2 (en)Selectable access rates in a memory and memory communication system
US11704270B2 (en)Networked computer with multiple embedded rings
KR102801506B1 (en) Nested rings on a toroidal computer network
CN116489064B (en) A method and device for constructing a multi-modal network test environment that combines internal and external fields
US20230280907A1 (en)Computer System Having Multiple Computer Devices Each with Routing Logic and Memory Controller and Multiple Computer Devices Each with Processing Circuitry
Zitouni et al.Communication architecture synthesis for multi-bus SoC
EP4487218A1 (en)A computer system
CN103425620B (en)The coupled structure of accelerator and processor based on multiple token-ring
Pham et al.Scalable low-latency inter-FPGA networks
CN107451427A (en)The computing system and accelerate platform that a kind of restructural gene compares
MirsadeghiImproving communication performance through topology and congestion awareness in HPC systems
Nunes et al.IPNoSys III: SDN Paradigm in a non-conventional NoC-based Processor
US12406118B1 (en)System and method for managing serial lanes in a multi-user emulation system

Legal Events

DateCodeTitleDescription
PB01Publication
PB01Publication
SE01Entry into force of request for substantive examination
SE01Entry into force of request for substantive examination
GR01Patent grant
GR01Patent grant

[8]ページ先頭

©2009-2025 Movatter.jp