DL380 G3 - Poor RAID 5i write performance

[John Pearce_4]

We have a DL380 G3 with 3x72GB 10K drives configured with RAID5, using the integrated 5i controller with the BBWCE installed and the default cache assignment of 50% read and 50% write.

We are running Smartstart installed W2k Server SP4 and using an old benchmark utility called nbench that writes 1GB to the drive and reads it back and gives us throughput.

The results in MB/Sec are
Write: 7.5
Read: 88.52

As a comparison, we also have a DELL 2850 with W2k Server using the DELL PERC4/DC with 128MB cache, 6x72GB 10K disks in a RAID 5 configuration with 2 drives on 1 channel and 4.

Using the same benchmark tool, in MB/Sec we are getting
Write: 65.0
Read: 55.0

As we are using a write intensive application we need to improve the DL380 performance, has anyone any ideas how to do this?

We have tried using the SA-642 instead of the 5i but this made no difference to the write performance although read performance improved slightly.

We did not have the cache on the 642 for this but the cache did not improve performance for the 5i, we assume because we have prolonged large writes that caching does not really benefit us in this case.

Is this poor disk performance the norm on the DL380 or are we missing something?
Any comments or suggestions would be appreciated.

John

[Akos Hegedus]

Dear John,

if your application is write intensive you don't have to configure RAID5. It is the slowest for write performance. You must use RAID1 and as much disks as possible.

I think if you get 3 more disk and make a new benchmark with 6 RAID5 configured disks you get better results then with the DELL box.

regards
Akos

[TwoProc]

John, I've got the same system, and I can say that the disk is ripping fast.

How 'bout... difference in Windows systems? Have you defragged them hard drives lately?

I see you've got 6 R5 drives on the other system ... brings up a thought - I've seen some Raid controllers have "sweet spots" and "dark spots" ... Maybe the lack of an optimal number of drives is messing you up? Maybe someone from HP/Compaq can comment on that ...

[John Pearce_4]

We have been asked by HP to use the cache enabler on the 642 and are going to try that.

This problem was on a newly installed system.

We do not think it is the source of the application problem we are having but it is an anomally that I need to clear up.

John

[Claas Kruhoefer]

hi,

your comparision is not that balanced.

You compare a drive set with more cache and more disks, with the DL380...

But anyway the performance values looks quite strange to me.

Is the disk space usage very high?
Did you have many random accesses with the tool or just sequential ?
Did you tried the tests after making defragment on the test volumes ?
did you have other DL380s with nearly this configuration to gain compareable values?


When the disk is already massively fragmented the benchmark would never come to write sequential, so it writes random access.

otherwise, maybe one Part is defect within your server.

[dfrank]

John - Have you been able to resolve this issue? I am also considering using some DL380s and am looking for RAID performance comparisons to the Dell 2850s that I am accustomed to.

Thanks!