Timing cached reads performance fall DL585

[Afanasiev]

Die All,

I updating DB server from DL385 to new DL585, I have performance fall.
Tests show:

HP DL385
hdparm -T /dev/cciss/c0d0p1

/dev/cciss/c0d0p1:
Timing cached reads: 6426 MB in 2.00 seconds = 3215.23 MB/sec
HP DL585
hdparm -T /dev/cciss/c0d0p1

/dev/cciss/c0d0p1:
Timing cached reads: 1844 MB in 2.00 seconds = 922.00 MB/sec

RAID state id OK
hpacucli controller all show config

Smart Array P400 in Slot 9 (sn: PAFGL0P9SWY1MQ)

array A (SAS, Unused Space: 0 MB)

logicaldrive 1 (33.9 GB, RAID 1+0, OK)

physicaldrive 1I:1:1 (port 1I:box 1:bay 1, SAS, 36 GB, OK)
physicaldrive 1I:1:2 (port 1I:box 1:bay 2, SAS, 36 GB, OK)

I changed various BIOS settings, kernel settings and kernels, but it haven’t good effect.

What can I do to correct it? Thanks!

[Vitaly Karasik_1]

Do you have the same type of disks (SAS) and the same RAID config in both cases?

[Afanasiev]

Yes, it's same disks

[loco_vikide]

You might want to compare the Read / Write Cache settings on your Raid controller (Model ???) on DL385 server and the SMart Array P400 controller on DL585 server

Cheers

[Ragu_3]

For the array type you are using, assuming that the controller is supported for GNU/Linux, changing your scheduler type can help.

[Mike Stroyan]

The performance of "hdparm -T" is not really a function of the disk or disk controller.
It measures transfer time from the RAM cache.
It is a measure of the RAM read rate.
A decrease in the reported "hdparm -T" rate may point to a problem with installing DIMMS in a proper high performance arrangement
Or it might be a side effect of NUMA vs. SUMA memory arrangement which will differ between a DL385 and larger systems like a DL585. You could check DIMM arrangement with dmidecode and experiment with changing the BIOS SUMA/NUMA setting.