|
108 | 108 | * code we determine the number of tapes M on the basis of workMem: we want |
109 | 109 | * workMem/M to be large enough that we read a fair amount of data each time |
110 | 110 | * we preread from a tape, so as to maintain the locality of access described |
111 | | - * above. Nonetheless, with large workMem we can have many tapes. |
| 111 | + * above. Nonetheless, with large workMem we can have many tapes (but not |
| 112 | + * too many -- see the comments in tuplesort_merge_order). |
112 | 113 | * |
113 | 114 | * |
114 | 115 | * Portions Copyright (c) 1996-2016, PostgreSQL Global Development Group |
@@ -247,6 +248,7 @@ typedef enum |
247 | 248 | * tape during a preread cycle (see discussion at top of file). |
248 | 249 | */ |
249 | 250 | #defineMINORDER6/* minimum merge order */ |
| 251 | +#defineMAXORDER500/* maximum merge order */ |
250 | 252 | #defineTAPE_BUFFER_OVERHEAD(BLCKSZ * 3) |
251 | 253 | #defineMERGE_BUFFER_SIZE(BLCKSZ * 32) |
252 | 254 |
|
@@ -2313,8 +2315,19 @@ tuplesort_merge_order(int64 allowedMem) |
2313 | 2315 | mOrder= (allowedMem-TAPE_BUFFER_OVERHEAD) / |
2314 | 2316 | (MERGE_BUFFER_SIZE+TAPE_BUFFER_OVERHEAD); |
2315 | 2317 |
|
2316 | | -/* Even in minimum memory, use at least a MINORDER merge */ |
| 2318 | +/* |
| 2319 | + * Even in minimum memory, use at least a MINORDER merge. On the other |
| 2320 | + * hand, even when we have lots of memory, do not use more than a MAXORDER |
| 2321 | + * merge. Tapes are pretty cheap, but they're not entirely free. Each |
| 2322 | + * additional tape reduces the amount of memory available to build runs, |
| 2323 | + * which in turn can cause the same sort to need more runs, which makes |
| 2324 | + * merging slower even if it can still be done in a single pass. Also, |
| 2325 | + * high order merges are quite slow due to CPU cache effects; it can be |
| 2326 | + * faster to pay the I/O cost of a polyphase merge than to perform a single |
| 2327 | + * merge pass across many hundreds of tapes. |
| 2328 | + */ |
2317 | 2329 | mOrder=Max(mOrder,MINORDER); |
| 2330 | +mOrder=Min(mOrder,MAXORDER); |
2318 | 2331 |
|
2319 | 2332 | returnmOrder; |
2320 | 2333 | } |
|