[Gluster-users] Slow read performance
Thomas Wakefield
twake at cola.iges.org
Thu Mar 7 20:08:13 UTC 2013
Everything is built as ext4, no options other than lazy_itable_init=1 when I built the filesystems.
Server mount example:
LABEL=disk2a /storage/disk2a ext4 defaults 0 0
Client mount:
fs-disk2:/shared /shared glusterfs defaults 0 0
Remember, the slow reads are only from gluster clients, the disks are really fast when I am local on the server testing the disks.
-Tom
On Mar 7, 2013, at 1:09 PM, Bryan Whitehead <driver at megahappy.net> wrote:
> Was just thinking, what is your mount options for your bricks (using inode64?)? Also, you are using xfs... right?
>
> When you created the filesystems did you allocate more inode space? -i size=512 ?
>
>
> On Thu, Mar 7, 2013 at 5:49 AM, Thomas Wakefield <twake at cola.iges.org> wrote:
> Still looking for help.
>
>
> On Mar 4, 2013, at 7:43 AM, Thomas Wakefield <twake at iges.org> wrote:
>
>> Also, I tested an NFS mount over the same 10GB link, and was able to pull almost 200MB/s. But Gluster is still much slower. Also I tested running it for a longer test, 105GB of data, and still showed that writing is MUCH faster. Which makes no sense when the disks can read 2x as fast as they can write.
>>
>> Any other thoughts?
>>
>> [root at cpu_crew1 ~]# dd if=/dev/zero of=/shared/working/benchmark/test.cpucrew1 bs=512k count=200000 ; dd if=/shared/working/benchmark/test.cpucrew1 of=/dev/null bs=512k
>> 200000+0 records in
>> 200000+0 records out
>> 104857600000 bytes (105 GB) copied, 159.135 seconds, 659 MB/s
>> 200000+0 records in
>> 200000+0 records out
>> 104857600000 bytes (105 GB) copied, 1916.87 seconds, 54.7 MB/s
>>
>>
>> On Mar 1, 2013, at 9:58 AM, Thomas Wakefield <twake at iges.org> wrote:
>>
>>> The max setting for performance.read-ahead-page-count is 16, which I did just try. No significant change.
>>>
>>> Any other setting options?
>>>
>>>
>>>
>>> On Feb 28, 2013, at 10:18 PM, Anand Avati <anand.avati at gmail.com> wrote:
>>>
>>>> Can you try "gluster volume set <volname> performance.read-ahead-page-count 64" or some value higher or lower?
>>>>
>>>> Avati
>>>>
>>>> On Thu, Feb 28, 2013 at 7:15 PM, Thomas Wakefield <twake at iges.org> wrote:
>>>> Good point, forgot to set a blcoksize, here are the redone dd tests:
>>>>
>>>> [root at cpu_crew1 ~]# dd if=/shared/working/benchmark/test.cpucrew1 of=/dev/null bs=128k
>>>> 40000+0 records in
>>>> 40000+0 records out
>>>> 5242880000 bytes (5.2 GB) copied, 65.4928 seconds, 80.1 MB/s
>>>> [root at cpu_crew1 ~]# dd if=/shared/working/benchmark/test.cpucrew1 of=/dev/null bs=1M
>>>> 5000+0 records in
>>>> 5000+0 records out
>>>> 5242880000 bytes (5.2 GB) copied, 49.0907 seconds, 107 MB/s
>>>> [root at cpu_crew1 ~]# dd if=/shared/working/benchmark/test.cpucrew1 of=/dev/null bs=4M
>>>> 1250+0 records in
>>>> 1250+0 records out
>>>> 5242880000 bytes (5.2 GB) copied, 44.5724 seconds, 118 MB/s
>>>>
>>>> Still not impressive.
>>>>
>>>> -Tom
>>>>
>>>>
>>>> On Feb 28, 2013, at 8:42 PM, Jeff Anderson-Lee <jonah at eecs.berkeley.edu> wrote:
>>>>
>>>>> Thomas,
>>>>>
>>>>> You have not specified a block size, so you are doing a huge number of small(ish) reads with associated round trips. What happens with dd bs=128k ..?
>>>>>
>>>>> Jeff Anderson-Lee
>>>>>
>>>>> On 2/28/2013 5:30 PM, Thomas Wakefield wrote:
>>>>>> Did a fresh dd test just to confirm, same results:
>>>>>>
>>>>>> [root at cpu_crew1 benchmark]# dd if=/dev/zero of=/shared/working/benchmark/test.cpucrew1 bs=512k count=10000
>>>>>> 10000+0 records in
>>>>>> 10000+0 records out
>>>>>> 5242880000 bytes (5.2 GB) copied, 7.43695 seconds, 705 MB/s
>>>>>> [root at cpu_crew1 benchmark]# dd if=/shared/working/benchmark/test.cpucrew1 of=/dev/null
>>>>>> 552126+0 records in
>>>>>> 552125+0 records out
>>>>>> 282688000 bytes (283 MB) copied, 37.8514 seconds, 7.5 MB/s
>>>>>>
>>>>>>
>>>>>>
>>>>>> On Feb 28, 2013, at 8:14 PM, Bryan Whitehead <driver at megahappy.net> wrote:
>>>>>>
>>>>>>> How are you doing the reading? Is this still an iozone benchmark?
>>>>>>>
>>>>>>> if you simply dd if=/glustermount/bigfile of=/dev/null, is the speed better?
>>>>>>>
>>>>>>>
>>>>>>> On Thu, Feb 28, 2013 at 5:05 PM, Thomas Wakefield <twake at iges.org> wrote:
>>>>>>> I get great speed locally, it's only when I add gluster in that it slows down. I get 2GB/s locally to the exact same brick. It's gluster that is having the read issue (80MB/s). But Gluster can write just fine, 800MB/s.
>>>>>>>
>>>>>>> The blockdev idea is a good one, and I have already done it. Thanks though.
>>>>>>>
>>>>>>> -Tom
>>>>>>>
>>>>>>> On Feb 28, 2013, at 7:53 PM, Ling Ho <ling at slac.stanford.edu> wrote:
>>>>>>>
>>>>>>>> Tom,
>>>>>>>>
>>>>>>>> What type of disks do you have? If they are raid 5 or 6, have you try setting the read-ahead size to 8192 or 16384 (blockdev --setra 8192 /dev/<sd?> ?
>>>>>>>>
>>>>>>>> ...
>>>>>>>> ling
>>>>>>>>
>>>>>>>> On 02/28/2013 04:23 PM, Thomas Wakefield wrote:
>>>>>>>>>
>>>>>>>>> Did anyone else have any ideas on performance tuning for reads?
>>>>>>>>>
>>>>>>>>> On Feb 27, 2013, at 9:29 PM, Thomas Wakefield <twake at iges.org> wrote:
>>>>>>>>>
>>>>>>>>>> Bryan-
>>>>>>>>>>
>>>>>>>>>> Yes I can write at 700-800MBytes/sec, but i can only read at 70-80 MBytes/sec. I would be very happy if I could get it to read at the same speed it can write at. And the 70-80 is sequential, not random for reads, same exact test commands on the disk server are in the 2+GB/s range, so I know the disk server can do it.
>>>>>>>>>>
>>>>>>>>>> -Tom
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> On Feb 27, 2013, at 7:41 PM, Bryan Whitehead <driver at megahappy.net> wrote:
>>>>>>>>>>
>>>>>>>>>>> Are your figures 700-800MByte/sec? Because that is probably as fast as your 10G nic cards are able to do. You can test that by trying to push a large amount of data over nc or ftp.
>>>>>>>>>>>
>>>>>>>>>>> Might want to try Infiniband. 40G cards are pretty routine.
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> On Wed, Feb 27, 2013 at 3:45 PM, Thomas Wakefield <twake at iges.org> wrote:
>>>>>>>>>>> I also get the same performance running iozone for large file sizes, iozone -u 1 -r 512k -s 2G -I -F.
>>>>>>>>>>>
>>>>>>>>>>> Large file IO is what I need the system to do. I am just shocked at the huge difference between local IO and gluster client IO. I know there should be some difference, but 10x is unacceptable.
>>>>>>>>>>>
>>>>>>>>>>> -Tom
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> On Feb 27, 2013, at 5:31 PM, Bryan Whitehead <driver at megahappy.net> wrote:
>>>>>>>>>>>
>>>>>>>>>>>> Every time you open/close a file or a directory you will have to wait for locks which take time. This is totally expected.
>>>>>>>>>>>>
>>>>>>>>>>>> Why don't you share what you want to do? iozone benchmarks look like crap but serving qcow2 files to qemu works fantastic for me. What are you doing? Make a benchmark that does that. If you are going to have many files with a wide variety of sizes glusterfs/fuse might not be what you are looking for.
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> On Wed, Feb 27, 2013 at 12:56 PM, Thomas Wakefield <twake at cola.iges.org> wrote:
>>>>>>>>>>>> I have tested everything, small and large files. I have used file sizes ranging from 128k up to multiple GB files. All the reads are bad.
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> Here is a fairly exhaustive iozone auto test:
>>>>>>>>>>>>
>>>>>>>>>>>> random random bkwd record stride
>>>>>>>>>>>> KB reclen write rewrite read reread read write read rewrite read fwrite frewrite fread freread
>>>>>>>>>>>> 64 4 40222 63492 26868 30060 1620 71037 1572 70570 31294 77096 72475 14736 13928
>>>>>>>>>>>> 64 8 99207 116366 13591 13513 3214 97690 3155 109978 28920 152018 158480 18936 17625
>>>>>>>>>>>> 64 16 230257 253766 25156 28713 10867 223732 8873 244297 54796 303383 312204 15062 13545
>>>>>>>>>>>> 64 32 255943 234481 5735102 7100397 11897 318502 13681 347801 24214 695778 528618 25838 28094
>>>>>>>>>>>> 64 64 214096 681644 6421025 7100397 27453 292156 28117 621657 27338 376062 512471 28569 32534
>>>>>>>>>>>> 128 4 74329 75468 26428 41089 1131 72857 1118 66976 1597 73778 78343 13351 13026
>>>>>>>>>>>> 128 8 100862 135170 24966 16734 2617 118966 2560 120406 39156 125121 146613 16177 16180
>>>>>>>>>>>> 128 16 115114 253983 28212 17854 5307 246180 5431 229843 47335 255920 271173 27256 24445
>>>>>>>>>>>> 128 32 256042 391360 39848 64258 11329 290230 9905 429563 38176 490380 463696 20917 19219
>>>>>>>>>>>> 128 64 248573 592699 4557257 6812590 19583 452366 29263 603357 42967 814915 692017 76327 37604
>>>>>>>>>>>> 128 128 921183 526444 5603747 5379161 45614 390222 65441 826202 41384 662962 1040839 78526 39023
>>>>>>>>>>>> 256 4 76212 77337 40295 32125 1289 71866 1261 64645 1436 57309 53048 23073 29550
>>>>>>>>>>>> 256 8 126922 141976 26237 25130 2566 128058 2565 138981 2985 125060 133603 22840 24955
>>>>>>>>>>>> 256 16 242883 263636 41850 24371 4902 250009 5290 248792 89353 243821 247303 26965 26199
>>>>>>>>>>>> 256 32 409074 439732 40101 39335 11953 436870 11209 430218 83743 409542 479390 30821 27750
>>>>>>>>>>>> 256 64 259935 571502 64840 71847 22537 617161 23383 392047 91852 672010 802614 41673 53111
>>>>>>>>>>>> 256 128 847597 812329 185517 83198 49383 708831 44668 794889 74267 1180188 1662639 54303 41018
>>>>>>>>>>>> 256 256 481324 709299 5217259 5320671 44668 719277 40954 808050 41302 790209 771473 62224 35754
>>>>>>>>>>>> 512 4 77667 75226 35102 29696 1337 66262 1451 67680 1413 69265 69142 42084 27897
>>>>>>>>>>>> 512 8 134311 144341 30144 24646 2102 134143 2209 134699 2296 108110 128616 25104 29123
>>>>>>>>>>>> 512 16 200085 248787 30235 25697 4196 247240 4179 256116 4768 250003 226436 32351 28455
>>>>>>>>>>>> 512 32 330341 439805 26440 39284 8744 457611 8006 424168 125953 425935 448813 27660 26951
>>>>>>>>>>>> 512 64 483906 733729 48747 41121 16032 555938 17424 587256 187343 366977 735740 41700 41548
>>>>>>>>>>>> 512 128 836636 907717 69359 94921 42443 761031 36828 964378 123165 651383 695697 58368 44459
>>>>>>>>>>>> 512 256 520879 860437 145534 135523 40267 847532 31585 663252 69696 1270846 1492545 48822 48092
>>>>>>>>>>>> 512 512 782951 973118 3099691 2942541 42328 871966 46218 911184 49791 953248 1036527 52723 48347
>>>>>>>>>>>> 1024 4 76218 69362 36431 28711 1137 66171 1174 68938 1125 70566 70845 34942 28914
>>>>>>>>>>>> 1024 8 126045 140524 37836 15664 2698 126000 2557 125566 2567 110858 127255 26764 27945
>>>>>>>>>>>> 1024 16 243398 261429 40238 23263 3987 246400 3882 260746 4093 236652 236874 31429 25076
>>>>>>>>>>>> 1024 32 383109 422076 41731 41605 8277 473441 7775 415261 8588 394765 407306 40089 28537
>>>>>>>>>>>> 1024 64 590145 619156 39623 53267 15051 722717 14624 753000 257294 597784 620946 38619 44073
>>>>>>>>>>>> 1024 128 1077836 1124099 56192 64916 36851 1102176 37198 1082454 281548 829175 792604 47975 51913
>>>>>>>>>>>> 1024 256 941918 1074331 72783 81450 26778 1099636 32395 1060013 183218 1024121 995171 44371 45448
>>>>>>>>>>>> 1024 512 697483 1130312 100324 114682 48215 1041758 41480 1058967 90156 994020 1563622 56328 46370
>>>>>>>>>>>> 1024 1024 931702 1087111 4609294 4199201 44191 949834 45594 970656 56674 933525 1075676 44876 46115
>>>>>>>>>>>> 2048 4 71438 67066 58319 38913 1147 44147 1043 42916 967 66416 67205 45953 96750
>>>>>>>>>>>> 2048 8 141926 134567 61101 55445 2596 77528 2564 80402 4258 124211 120747 53888 100337
>>>>>>>>>>>> 2048 16 254344 255585 71550 74500 5410 139365 5201 141484 5171 205521 213113 67048 57304
>>>>>>>>>>>> 2048 32 397833 411261 56676 80027 10440 260034 10126 230238 10814 391665 383379 79333 60877
>>>>>>>>>>>> 2048 64 595167 687205 64262 87327 20772 456430 19960 477064 23190 540220 563096 86812 92565
>>>>>>>>>>>> 2048 128 833585 933403 121926 118621 37700 690020 37575 733254 567449 712337 734006 92011 104934
>>>>>>>>>>>> 2048 256 799003 949499 143688 125659 40871 892757 37977 880494 458281 836263 901375 131332 110237
>>>>>>>>>>>> 2048 512 979936 1040724 120896 138013 54381 859783 48721 780491 279203 1068824 1087085 97886 98078
>>>>>>>>>>>> 2048 1024 901754 987938 53352 53043 72727 1054522 68269 992275 181253 1309480 1524983 121600 95585
>>>>>>>>>>>> 2048 2048 831890 1021540 4257067 3302797 75672 984203 80181 826209 94278 966920 1027159 111832 105921
>>>>>>>>>>>> 4096 4 66195 67316 62171 74785 1328 28963 1329 26397 1223 71470 69317 55903 84915
>>>>>>>>>>>> 4096 8 122221 120057 90537 60958 2598 47312 2468 59783 2640 128674 127872 41285 40422
>>>>>>>>>>>> 4096 16 238321 239251 29336 32121 4153 89262 3986 96930 4608 229970 237108 55039 56983
>>>>>>>>>>>> 4096 32 417110 421356 30974 50000 8382 156676 7886 153841 7900 359585 367288 26611 25952
>>>>>>>>>>>> 4096 64 648008 668066 32193 29389 14830 273265 14822 282211 19653 581898 620798 51281 50218
>>>>>>>>>>>> 4096 128 779422 848564 55594 60253 37108 451296 35908 491361 37567 738163 728059 67681 66440
>>>>>>>>>>>> 4096 256 865623 886986 71368 63947 44255 645961 42689 719491 736707 819696 837641 57059 60347
>>>>>>>>>>>> 4096 512 852099 889650 68870 73891 31185 845224 30259 830153 392334 910442 961983 60083 55558
>>>>>>>>>>>> 4096 1024 710357 867810 29377 29522 49954 846640 43665 926298 213677 986226 1115445 55130 59205
>>>>>>>>>>>> 4096 2048 826479 908420 43191 42075 59684 904022 58601 855664 115105 1418322 1524415 60548 66066
>>>>>>>>>>>> 4096 4096 793351 855111 3232454 3673419 66018 861413 48833 847852 45914 852268 842075 42980 48374
>>>>>>>>>>>> 8192 4 67340 69421 42198 31740 994 23251 1166 16813 837 73827 73126 25169 29610
>>>>>>>>>>>> 8192 8 137150 125622 29131 36439 2051 44342 1988 48930 2315 134183 135367 31080 33573
>>>>>>>>>>>> 8192 16 237366 220826 24810 26584 3576 88004 3769 78717 4289 233751 235355 23302 28742
>>>>>>>>>>>> 8192 32 457447 454404 31594 27750 8141 142022 7846 143984 9322 353147 396188 34203 33265
>>>>>>>>>>>> 8192 64 670645 655259 28630 23255 16669 237476 16965 244968 15607 590365 575320 49998 43305
>>>>>>>>>>>> 8192 128 658676 760982 44197 47802 28693 379523 26614 378328 27184 720997 702038 51707 49733
>>>>>>>>>>>> 8192 256 643370 698683 56233 63165 28846 543952 27745 576739 44014 701007 725534 59611 58985
>>>>>>>>>>>> 8192 512 696884 776793 67258 52705 18711 698854 21004 694124 621695 784812 773331 43101 47659
>>>>>>>>>>>> 8192 1024 729664 810451 15470 15875 31318 801490 38123 812944 301222 804323 832765 54308 53376
>>>>>>>>>>>> 8192 2048 749217 68757 21914 22667 48971 783309 48132 782738 172848 907408 929324 51156 50565
>>>>>>>>>>>> 8192 4096 707677 763960 32063 31928 47809 751692 49560 786339 93445 1046761 1297876 48037 51680
>>>>>>>>>>>> 8192 8192 623817 746288 2815955 3137358 48722 741633 35428 753787 49626 803683 823800 48977 52895
>>>>>>>>>>>> 16384 4 72372 73651 34471 30788 960 23610 903 22316 891 71445 71138 56451 55129
>>>>>>>>>>>> 16384 8 137920 141704 50830 33857 1935 41934 2275 35588 3608 130757 137801 51621 48525
>>>>>>>>>>>> 16384 16 245369 242460 41808 29770 3605 75682 4355 75315 4767 241100 239693 53263 30785
>>>>>>>>>>>> 16384 32 448877 433956 31846 35010 7973 118181 8819 112703 8177 381734 391651 57749 63417
>>>>>>>>>>>> 16384 64 710831 700712 66792 68864 20176 209806 19034 207852 21255 589503 601379 104567 105162
>>>>>>>>>>>> 16384 128 836901 860867 104226 100373 40899 358865 40946 360562 39415 675968 691538 96086 105695
>>>>>>>>>>>> 16384 256 798081 828146 107103 120433 39084 595325 39050 593110 56925 763466 797859 109645 113414
>>>>>>>>>>>> 16384 512 810851 843931 113564 106202 35111 714831 46244 745947 53636 802902 760172 110492 100879
>>>>>>>>>>>> 16384 1024 726399 820219 22106 22987 53087 749053 54781 777705 1075341 772686 809723 100349 96619
>>>>>>>>>>>> 16384 2048 807772 856458 23920 23617 66320 829576 72105 740848 656379 864539 835446 93499 101714
>>>>>>>>>>>> 16384 4096 797470 840596 27270 28132 88784 834938 83428 836809 342898 953274 969699 96499 100657
>>>>>>>>>>>> 16384 8192 760063 812942 41022 40014 102020 812135 104063 820346 229870 1252409 1398306 100959 103793
> ...
>
> [Message clipped]
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://supercolony.gluster.org/mailman/listinfo/gluster-users
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://supercolony.gluster.org/pipermail/gluster-users/attachments/20130307/228289fa/attachment.html>
More information about the Gluster-users
mailing list