Different HDD in one RAID

PiotrIr

Hi

I have a performance problem on one of my servers. After taking counter logs I noticed that average disk queue is very high.
I looked and server and noticed that one of HDD has different part number (however both are suitable for this server) so I wonder if it may be a problem. Does somebody experienced something similar?

Server: HP DL380 G4
RAID 5 -3 HDD (2x 404708-001 1x 289044-001)

Many thanks for your help

jibbajabba

Different part number might mean different firmwares ..

We had performance issues (using Adaptec raid though) when we used SAS disks where one disk had a different firmware ....

But those are SCSI so I don't honestly know ..

jibbajabba

Ah I see, 289044-001 is the Retail Part Number where 404708-001 is the Spare Part Number. So there isn't probably a technical difference between the drives itself, apart from firmware (iff applicable). On Seagate drives the firmware is part of the product number - might be worth checking .. For example, if the product number (not model number) ends in '-001' or similar - it means it uses firmware '001'. If that is the same with HP disks, then it might be ok .. You say average disk queue - don't forget - Raid5 has 'rubbish' write speed anyway ...

PiotrIr

Many thanks for your reply.

On Compaq StorageWorks 4000 RAID Array- Notice: (Revision) HP ProLiant Server and StorageWorks Hard Drive Model-to-Spare Part Number Matrix - c00872817 - Centrum Wsparcia Biznesowego HP
I can see that both HDD have the same Option Part number and different Spare Part number. Differences it that one is RoHS compliant (any idea what is it?) and second not.

About RAID 5 I agree that it is rubbish. However when you see average Disk Queue Length – 8.5 (min 0.747 max 34.060) it can’t be normal. As far as I know under 2 you have bottleneck.

Once again many thanks.

jibbajabba

Means it has no hazardous material according to EU laws.

RoHS

PiotrIr

Many thanks.

Do you have any interesting article how to troubleshooting problems like that? I can’t exclude that it is not hardware issue. I wonder how to do performance test to check if it is hardware or if it is software related problem.

This server is Lotus notes and file server.

Claymoore

PiotrIr wrote: »

This server is Lotus notes and file server.

2 completely different I/O profiles - 3 if you use transaction logging on the Notes database. I don't know much about Notes, but I do know a bit about Exchange and storage, and this setup is going to give you terrible throughput regardless of firmware. You have log files, a database and shared files on the same spindles and they all want to use the disk differently.

Notes database - completely random reads and writes, noticeable write penalty due to parity calculation on RAID 5, RAID 1 or RAID 10 preffered
Transaction logs - sequential writes, crippling write penalty on RAID 5, RAID 1 almost required.
File Server - random read writes but not as intensive as a db, RAID 5 works fine and is preferred.

You need to separate the logs from the DB for two reasons. One, so that if you lose the physical disks where the DB is located you can restore and replay the transaction logs which are safe on other disks. Two, log files are sequential write/read operations and keeping them separate keeps the disk write heads in the same location so the disk seek time and rotational latency don't slow you down. You start adding fragmentation from a file server and now the log file is written all over the disk. Your disks simply can't keep up with all the requests scattered all over the disk and the queue is building up. So, to answer your question, the problem is both hardware and software.

PiotrIr

Many thanks for your reply.

I’m also Microsoft guy and I was asked to troubleshoot issue on unfortunately lotus notes. I know best practices for Exchange however lotus notes is completely deferent.
First thing which I noticed is that every user has deferent database file and looks that without any logs. Something like keeping .pst file on shared drive in the network.

For me it looks like solution from 19 century but maybe I’m wrong. Anyway all together files are 15GB so quite small. It is only 25 mailboxes and the biggest is 1.9GB. With these parameters in my opinion system shouldn’t have performance issues on any RAID and disk configuration especially that problem occurred only month ago and they didn’t have any performance issues for 3 years.

I more believe that it is hardware problem but I need to improve this somehow before start to change it.
Could you help me find a method how to do this please?

tiersten

RAID 5 has bad performance with a low number of spindles. You also need to choose the correct stripe size for the workload.

UnixGuy

Well, general answer is yes definitely.

Even if the disks are same size, but from different vendors, then you might encounter performance issues, because sizes of cylinders will differ between two disks, and depends on type of I/O you have, you might get some performance issues.

For RAID 5, if sizes are different, then you will lose space because the total size will be multiple of the size of smallest disk.

to mitigate the problem, check these things:

1. Application recommendation for volume "stripe unit" (chunk size). PS: to modify this you will have to destroy the volume, build it again, and restore from back, but sometimes it can be the real performance bottleneck.

2. Check disks firmwares, and RAID controller firmware if there are possible upgrades.

3. If you can afford having all disks identical and from same vendor, do that.

4. If you can afford RAID 1+0 and you can justify the cost, you will have tangible performance enhancement.

PiotrIr

Mamy thanks for your help.