RAID 5 performance problem

[vbourbeau]

Hi

I'm not very good in storage but I have a performance problem and I don't know where to look...

I have 2 x Esx 4 server, twice are HP Proliant dl380 G5 with local storage

first have 8 x 146 g in raid 5 and I can got 70000 Kbps

Second have 4 x 300 g in raid 5 and I can only got 10000 Kbps

The first one is 7 time faster why?

[f10]

Do we have a battery backed read/write cache installed on the other HP proliant server ?

We would need more data to diagnose this issue, use esxtop command from the console. I would recommend that you go through KB http://kb.vmware.com/kb/1008205 to identify storage issues.

NOTE : All arrays perform differently, but DAVG/cmd, KAVG/cmd, and GAVG/cmd should not exceed more than 10 milliseconds (ms). These values should not exceed 10 ms for a sustained period of time.

Also monitor the /var/log/vmkernel and check if there are any error that keep repeating

Hope this information helps :smileyhappy:

If you found this or other information useful, please consider awarding points for "Correct" or "Helpful".

[vbourbeau]

Both is a HP P400 smart array the faster one have a battery but not the slow one (can this be the problem?)

Esxtop Slow one

ADAPTR PATH CMDS/s READS/s WRITES/s MBREAD/s MBWRTN/s DAVG/cmd KAVG/cmd GAVG/cmd QAVG/cmd

vmhba0 - 115.07 47.03 68.04 0.37 8.26 18.40 0.01 18.41 0.05

Esxtop Fast one

ADAPTR CMDS/s READS/s WRITES/s MBREAD/s MBWRTN/s DAVG/cmd KAVG/cmd GAVG/cmd QAVG/cmd

vmhba0 3305.44 2961.16 342.37 23.30 52.23 0.23 0.01 0.24 0.01

In the log I have this line:

Jul 22 12:51:45 carbone vmkernel: 1:01:46:47.530 cpu5:4301)WARNING: LinuxFileIoctl: 819: unsupported: (obj=0x4100bae7b130/1 cmd=0x2285, userData=0x44c50d8)

Thank

[DSTAVERT]

Without a Battery the array can not do write caching. Without write caching you will have a huge performance penalty. Get the battery and enable write caching.

[f10]

Hi,

I dont want to jump onto conclusions but I has a similar issue where we had extremely poor write performance without the batter backed cache installed on the controller, however this was for a Windows Server but the underlying storage technology remains the same for ESX.

From esxtop I see that the DAVG/cmd on the slow box is 18, if it remains about 10 constantly then there is a latency from the storage.

I would recommend that you install the BBWC on the SMA controller else swap it with the good known hosts, also check the RPM for the hard drives that are being used.

Hope this information helps :smileyhappy:

If you found this or other information useful, please consider awarding points for "Correct" or "Helpful".

[vbourbeau]

OK I will look to buy the bbwc, for now I can't swap it because the faster one is in production.

I will give you the result since I've got the battery

Do you know what this line mean

Jul 22 12:55:37 carbone vmkernel: 1:01:50:39.530 cpu0:4301)WARNING: LinuxFileIoctl: 819: unsupported: (obj=0x4100bae7b130/1 cmd=0x2285, userData=0x44c50d8)

[vbourbeau]

I look the card and it miss only the battery kit... I'm not sure, but the battery kit is only use when the server goes power failure?

The card seems have cache, it have a kind of RAM attatch to it.

[DSTAVERT]

The battery allows you to use write caching. There is a huge performance penalty without write caching enabled. I don't believe you can enable write caching on those controllers without the battery. Without a battery and write caching enabled you would trash your vmdks if the power went out or you did a hard power down.

[vbourbeau]

The battery and the write caching solve my problem...

Thank

[DSTAVERT]

Excellent.