Date: Tue, 15 May 2007 16:02:14 -0500 From: Eric Anderson <anderson@freebsd.org> To: freebsd-performance@freebsd.org Subject: Re: PERC5i throughput [was: possible issues with Dell/Perc5 raid] Message-ID: <464A1FD6.8070104@freebsd.org> In-Reply-To: <000301c7970e$6dbdcc80$9501a8c0@skylinecorp.net> References: <Pine.BSF.4.64.0705101545340.34925@tdream.lly.earlham.edu> <464474F0.3040306@tomjudge.com> <7579f7fb0705110735h1a65ef7atcab00bdbc25224d6@mail.gmail.com> <4649A71C.4030302@skylinecorp.com> <Pine.BSF.4.64.0705150948390.61298@tdream.lly.earlham.edu> <4649CC6B.1050203@tomjudge.com> <000301c7970e$6dbdcc80$9501a8c0@skylinecorp.net>
next in thread | previous in thread | raw e-mail | index | archive | help
On 05/15/07 11:30, Kevin Kobb wrote:
> Tom Judge wrote:
>> Randy Schultz wrote:
>>> On Tue, 15 May 2007, Kevin Kobb spaketh thusly:
>>>
>>> -}These reports on poor performance using mpt seem to be on SATA
>>> -}drives. Has anybody been seeing this using SAS drives? -}
>>> -}We are testing Dell PE840s with hot swap SAS drives, and seem to
>>> get
>>> decent -}performance, though I haven't run any benchmarks yet. Any
>>> opinions if the -}PERC5i/mfi is a better choice than the SAS5iR/mpt
>>> combination?
>>>
>>> Would it be possible for you to install blogbench and give it a quick
>>> run while your system is idle? We will be ordering a new box in 1-2
>>> months that we need high disk I/O capabilities. We've been looking
>>> at
>>> the PERC5i set up RAID 5 with 4-6 SAS drives. I'ld be interested in
>>> what you're seeing for throughput on the PERC5i.
>>>
>> I have attached some blogbench tests from the following configs:
>>
>> Perc5i 4 * SAS 15K RPM 146 Gig disks in raid 5.
>>
>>
>> Perc5e 15 * SATA 7.2K RPM 500 Gig disks in raid 50 (3 raid 5 volumes
>> 5 disks each)
>>
>>
>> Other configurations that I may be able to test are 2 SATA in raid 1
>> and 2 SAS in raid 1 (both on perc5i).
>
> Ok, I installed blogbench (which I have not used before) and ran a
> couple of quick tests.
> I have a SAS 5iR controller, (not a PERC5 though) on a PE840, with 2 GB
> RAM, and 2 146 GB 10K RPM hot plug drives in a RAID 1.
>
> I ran the test and noticed:
> "mpt0: QUEUE FULL EVENT: Bus 0x00 Target 0x20 Depth 121" messages in my
> logs.
> Then I ran, camcontrol tags 0:0:0 -N 119 -v, and ran it again.
>
> I didn't get any messages this time and got the results indicated in
> test 2.
Just for an additional data point, here's what a 2Gb Fiber channel
connected array with 16 750GB SATA disks looks like:
Frequency = 10 secs
Scratch dir = [/vol3/test/]
Spawning 3 writers...
Spawning 1 rewriters...
Spawning 5 commenters...
Spawning 100 readers...
Benchmarking for 30 iterations.
The test will run during 5 minutes.
Nb blogs R articles W articles R pictures W pictures R
comments W comments
48 85428 2630 60484 2865
46462 8899
78 77246 1670 56118 1547
48798 5714
101 68730 1634 47639 1103
41230 5108
127 64230 1663 43522 1422
35531 4517
150 64072 1326 42968 1330
35485 4165
168 39332 1163 26511 993
20236 2697
194 53474 1527 35969 1137
32142 4251
215 55310 1362 37140 1274
30882 4401
232 49766 1203 32995 1046
29133 3979
251 38767 1061 27122 909
23652 3130
272 40820 1344 29009 920
23557 3728
285 23580 771 15778 746
14036 2406
300 26545 979 18758 853
14721 3001
323 34491 1319 23422 1222
20155 4197
333 15418 732 10068 738
7867 2324
361 21929 1340 15030 1505
12914 5008
373 13122 524 9329 743
7842 2145
396 24737 871 17423 1177
14929 3182
404 8858 323 6651 345
5864 1165
433 25465 1450 18236 1472
15915 3982
438 9460 379 6477 259
4627 1284
454 15580 869 10450 1087
9702 3297
470 11886 874 8590 603
7578 2541
487 16931 1088 11651 803
9846 3231
496 11122 559 7517 580
5341 2065
517 13609 883 9288 1150
8291 3611
545 8043 1610 5581 1022
4973 3771
557 5049 686 3784 542
2195 2383
578 9534 1451 6703 1334
6541 5202
593 4383 834 3230 1011
3432 1381
Final score for writes: 593
Final score for reads : 18671
My system also is pretty busy doing an rsync to/from some other arrays,
so the numbers are lower that what I'd get on an idle system.
I'll try this again on a faster system if I get the chance.
Eric
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?464A1FD6.8070104>