jemalloc performance for very small allocations

Fri Apr 26 09:39:10 PDT 2013

Hi,
 I am using jemalloc 3.1.0, compiled with gcc 4.1.2 on CentOS release 5.6 for a compute intensive multithreaded application where bulk of allocations falls in 8-byte, 16-byte (stats below). We are seeing very high runtime in tcache_alloc_small_hard(), arena_tcache_fill_small() and arena_bin_malloc_hard().  The 8 and 16 bytes allocations happen in very large number. We probably want the allocations to come form tcache_alloc_easy() to not hit the locks and take less runtime.  The allocations are mostly temporary in nature -- especially 8-byte ones -- so some number of 8-byte allocations are done, then they are freed and then alloc/free process repeated. 

The runs use 8 threads (so 48 arenas and all other jemalloc settings are default) and are on an Intel Xeon server with 12-cores (no hyperthreading and no other user so no contention with other user; the peak memory in application is ~6.5GB and server has 128GB memory so no memory shortage/swapping etc.); the jemalloc output at the end of run is given below. As we can see, pretty much 8 and 16 byte bins are overloaded. 

A run with TCACHE_NSLOTS_SMALL_MAX=10000, LG_TCACHE_MAXCLASS_DEFAULT=10, changing small bins to hold only upto 224 bytes (NBINS=15),  and hardcoding 
tcache_bin_info[i].ncached_max = TCACHE_NSLOTS_SMALL_MAX in tcache.c file improves runtime by about 16% (for overall application runtime) but increases peak memory from 6.5GB to 6.7GB. We want the runtime to at least improve by another 10% without preferably any increase in peak memory (so even 6.5GB to 6.7GB is not desirable). 
Any suggestions on what changes to jemalloc settings to try?
thanks a lot!,
--Rajat

___ Begin jemalloc statistics ___
Version: 3.1.0-0-g3b1f3aca54fede23299cde9034f7b909c3d290d7
Assertions disabled
Run-time option settings:
  opt.abort: false
  opt.lg_chunk: 22
  opt.dss: "secondary"
  opt.narenas: 48
  opt.lg_dirty_mult: 5
  opt.stats_print: false
  opt.junk: false
  opt.quarantine: 0
  opt.redzone: false
  opt.zero: false
  opt.tcache: true
  opt.lg_tcache_max: 15
CPUs: 12
Arenas: 48
Pointer size: 8
Quantum size: 16
Page size: 4096
Min active:dirty page ratio per arena: 32:1
Maximum thread-cached size class: 32768
Chunk size: 4194304 (2^22)
Allocated: 1209815024, active: 1295314944, mapped: 1832910848
Current active ceiling: 1321205760
chunks: nchunks   highchunks    curchunks
           446          443          437
huge: nmalloc      ndalloc    allocated
            0            0            0

Merged arenas stats:
assigned threads: 1
dss allocation precedence: N/A
dirty pages: 316239:9435 active:dirty, 1133 sweeps, 43514 madvises, 198766 purged
            allocated      nmalloc      ndalloc    nrequests
small:     1048219632  14499926892  14459296188  16587774427
large:      161595392         1658         1282         2659
total:     1209815024  14499928550  14459297470  16587777086
active:    1295314944
mapped:    1828716544
bins:     bin  size regs pgs    allocated      nmalloc      ndalloc    nrequests       nfills     nflushes      newruns       reruns      curruns
            0     8  501   1     17659680   7173034900   7170827440   8020202160     71750873     71709910       282770     17873119         4463
            1    16  252   1    399942864   7263860606   7238864177   8275464931     72760253     72395028       138537   1394609632       105755
            2    32  126   1    340674496     34841421     24195343    229801652       504480       257943       128506      1157771        89699
            3    48   84   1     78136368      2304192       676351     11567784       101000        18024        22085        18898        19409
            4    64   63   1      1573312      3402405      3377822      5487283       217389        67666        33203       121936          454
            5    80   50   1     17949200      1591440      1367075      2726511       177324        40895        23869        56202         4552
            6    96   84   2     25039488      7866030      7605202     19022100       150183        96444        40254       127493         3556
            7   112   72   2     25960928      4878810      4647016      6646579       277346        80409        39762        81691         3309
            8   128   63   2      4877696       345189       307082       887445        91917        17364         2600        18459          606
            9   160   51   2      3152000       470435       450735       712418        81161        19755         6687        13692          388
           10   192   63   3      7931712      1356246      1314935      2595825        81608        31104        18297        14202          792
           11   224   72   4       583072      1666870      1664267      4112733        94788        84418        18487        14044           37
           12   256   63   4      4441088       214118       196770       531195        50982        23215         2773         3298          332
           13   320   63   5       161600      1507044      1506539      3623214       112592        45548        16571        22779            9
           14   384   63   6     63520128      1115114       949697      1845693        61670        33546        15219         8725         3117
           15   448   63   7     56578368       922862       796571       908379        69788        43446        13528         7001         2006
           ...<cut>...
large:   size pages      nmalloc      ndalloc    nrequests      curruns
         4096     1          413          413          956            0
         8192     2          378          378          821            0
        12288     3           13           13           27            0
        16384     4            2            2            3            0
        20480     5            3            3            3            0
        ...<cut>...

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://jemalloc.net/mailman/jemalloc-discuss/attachments/20130426/81539833/attachment.html>