DL385 I/O performance issue

[Bob Pace]

I have a customer that is have I/O performace issues with a DL385

Here is the info and I have attached data for the DL380G4 and the DL385
-----------------------

First, I want to thank all who tried to help with the problems I am experiencing with the DL385 G1 just purchased. The disk performance of this server is less than poor. I am including screenshots and specs for the server and other servers for comparison.



DL385 G1 w/ 2 dual core processors (2.4 GHz), 16 GB memory running W2K3 x64 Standard Edition. Server has 6 â 146GB 15k rpm drives configured RAID5. System board was replaced by HP Tech on Monday. See DL385_Benchmark.png



(2) DL145 G2 w/ 2 dual core processors (2.2 GHz), 4 GB memory running W2K3 x64 Standard Edition. Server has 2 â 36GB 15k rpm drives configured RAID1 (software). See DL145_1Benchmark.png and DL145_2Benchmark.png



DL380 G4 w/ 2 Xeon 3.6GHz processors, 8 GB memory running W2K3 (32-bit) Standard Edition. Server has Server has 6 â 146GB 15k rpm drives configured RAID5 (locally). See DL380_Benchmark.png



The DL380 and DL385 are using the integrated 6i controller for the local drives. Both are using the Compaq Softpaq software/drivers. All servers are built using the same procedures.

[Tony Davidson]

Bob, I noticed that the access time is faster on the 385, but the xfer rate is slower. I would like to make sure the 6i controller settings are the same read / write 50% etc...
Another thing, we bought a dl385 and the battery was dead on the controller until we flashed the firmware. Just a couple of thoughts.

[Bob Pace]

Hi

Thanks for the input. The customers states that neither server has the BBWC module installed.

[ispcolohost]

I haven't run Windows or raid 5 on these systems but maybe it's a driver or raid 5 issue with this particular raid controller? I'm running RedHat Enterprise 4 on my DL385's and with the classis Bonnie disk benchmark program I am seeing over 85 MB/sec sustained writes and 120+ MB/sec reads on a pair of mirrored 10krpm drives. I would have expected having six spindles in raid 5 to get around the write penalty but maybe the controller itself is the limiting factor.

David