New IBM server slow disks!!! | SQL Server Performance Forums

SQL Server Performance Forum – Threads Archive

New IBM server slow disks!!!

Hi, I have just setup a new IBM x346 server where I have a serveraid 6M controller card (with 128mb of cache) I have 8 disks on the rack (8 * 36gb 15krpm)
I have 2 channels connected between the rack and the controller. I have 4 disks in raid5 (2 disks on each channel)
2 in raid 0 (1 disk on each channel)
and 2 in raid 0 (1 disk on each channel) but the disk throughput is slow with 50% read+ 50% writes!!!! (iometer used) 1 local disk has a throughput of 12mb/s
a raid 0 on the rack go up to 17mb/s
the raid 5 = 15mb/sec 4 disks in raid 0 go to 22mb/s!!! if I test 100% read only, then I go to :
1 local disk= 80mb/s
raid 0 = 80mb/s
raid 5 = 25mb/s wow… the raid 5 don’t change… the 2 disks raid 0 don’t provide any advantage against a single disk!!! if I read (100% read) my raid 5 + raid 0 + raid 0 at the same time the throughput = 100mb/s !!!!! while I expect : 80 + 80 + 25 = 185mb/s!!! I don’t understand whats appends…
specially with the raid 0 config.
if the 2 disks of the raid 0 are on the same channel, the performance slow down to 8mb/s half of the speed of 1 disk!!! does the dual channel link is the problem?
all the drivers & firmware are uptodate thanks for your help. jerome. to check for more information. Satya SKJ
This posting is provided “AS IS” with no rights for the sake of knowledge sharing.
But this doesn’t help me in the way of:
when I add disks in a Raid 0 the performance is not improved and, sometimes, is degraded compared to a single disk!!! for example, a 100% read test demonstrate that a single disk go to 80mb/s and a raid 0 go to… 80mb/s while 4 disks in raid 5 go to 25mb/s!!!!! do you think its good????

what test are you doing?
only look at MB/sec when doing sequential
if random, then don’t bother looking at MB/sec, just look at IOPS, example: 8K random IO
expect 200 IOPS at moderate queue depth
this is 1.6MB/sec per disk
well… after some tests and changing in the config I have found a good solution. now I’m using the SQLIOStress tool, which simulate an SQL Server load.
I have changed my raid 0 to a raid 00, a spanned array configuration. (finally its the same result has a raid 0)
the test jump from 4mb/s to 31mb/s for the raid!!!
my raid 5 go to 27mb/s
and a simple disk go to 8mb/s. this 31mb/s on 2 disks outperform another server where I have 4 disks in raid 10 on a SAN (result = 22mb/s) Now I’m testing when I access the raid 5 & raid 00 at the same time. the usage is a small data warehouse.
I’ll put the tempdb on the raid 00 and the DW himself on the raid 5
and the other raid 00 will host staging & temporary databases

Well… after a lot of tests and changes in the config… writing access is good on the raid disks, but the read access is poor!!!
with iometer… Setup: 1 local disk no raid, no cache
8kb, 50% random, 100% read -> 350 io/sec; 3mb/s
8kb, 50% random, 100% write -> 680 io/sec; 5.5mb/s Setup: 4 disks in raid 10, 128 mb cache
8kb, 50% random, 100% read -> 250 io/sec; 2mb/s
8kb, 50% random, 100% write -> 1200 io/sec; 9.5mb/s writing process is excellent (twice the performance of 1 disk)
but reading process…. no comment… which parameter could impact the reading step?

Are you using the RAID controller card that came with the server willgart? MeanOldDBA
[email protected] When life gives you a lemon, fire the DBA.
its a second controller.
there is an internal default controller, and we have added a second one to connect the external disks.
its a ServeRaid 6M controller. (IBM Server) we have applied some changes in the hardware config and I’ll test this.