Hi! On Apr 5, 2005 5:23 AM, John Andersen wrote:

On Monday 04 April 2005 12:35 am, Hugo wrote:

...

Hello!

Situation: a dual opteron 9.2 server with SATA-RAID (3Ware - driver: 3w-9xxx) and a dual opteron 9.0 accessing the RAID with NFS. I'm moving data from the "client" (9.0) to the server with mv command. Load on client is nothing. Load on server is 2.5-3 and yet there is nothing running there; Top has nfsd process at the top with 1-2 % CPU. Computers are connected via Gigabit ethernet. Data transfer rate is very slow: 2.2Mb/s!

The problem is not at the client end as we had the same kind of results transferring data from another (RH) server to the new SuSE 9.2.

Installed kernel is 2.6.8-24.11 (default SMP-kernel from SuSE). The RAID driver is also from SuSE 9.2 (I could not find information about which.

Any ideas? Any pointers to check? We really do not have any idea what is slowing down the server?

-- HG

Are you commited to that controller? Do you have the option of jumpering it to be simple sata controller(s) without using the raid portion, and then using software raid?

No, not easily. I have lot's of data there... BTW, I'm using LVM in between the RAID and the partitions - but that should not make any difference, right? -- Hugo.

Hi! On Apr 5, 2005 9:02 AM, John Andersen wrote:

On Monday 04 April 2005 09:53 pm, Hugo wrote:

...

...

Are you commited to that controller? Do you have the option of jumpering it to be simple sata controller(s) without using the raid portion, and then using software raid?

No, not easily. I have lot's of data there... BTW, I'm using LVM in between the RAID and the partitions - but that should not make any difference, right?

Should not make a difference as i understand it, although it should not be necessary. Are you sure the raid is not degraded (disk went bad)? Thats too obvious to miss, so I assume its not the case.

The RAID is out only "hard disk". So some part of it is reserved for boot, but rest of the partitions are on LVM so that we can adjust the sizes as needed later on. No, the hard disks seem fine. -- Hugo.

[8<]

...

The RAID is out only "hard disk". So some part of it is reserved for boot, but rest of the partitions are on LVM so that we can adjust the sizes as needed later on.

No, the hard disks seem fine.

I have situation which is completely different, uses different hardware and disks, but the same effect as you: Write access to my RAID array is about 2mb/s, sequential reads 22mb/s, random reads about 11-12 mb/s.

I'm using SuSE 9.2 pro with a Compaq SmartArray 2P / 3200 SCSI RAID controller and a stack of 8 SCSI drives, 18 GB 10,000 RPM. I boot from a separate IDE drive. The raid array is set up as three ReiserFS filesystems each in its own logical volume.

I can't explain it.

Are you using the SA controller in a Compaq box? If not, there's your problem when writing to disk. Reads are generally fine, but writes are disasters in that config. Anders.

On Tuesday 05 April 2005 10:07 am, Tom Peters wrote:

I did finally install a daemon that writes events to the messages log file generated my the RAID controller. But that's specific to the Compaq  controller.

And this did WHAT for you? Did it help explain the problem? You left us guessing.... -- _____________________________________ John Andersen

Hi! On Apr 6, 2005 9:57 PM, Carlos E. R. wrote:

The Monday 2005-04-04 at 11:35 +0300, Hugo wrote:

...

Any ideas? Any pointers to check? We really do not have any idea what is slowing down the server?

Just one idea. You can check, locally, the transfer speed of the raid setup. Perhaps you can also setup a non-raid partition, export it through nfs, and check the speed from the client. Then you could know what is the problem: one, two, or the combination.

The problem is local (did some runs with bonnie++). And we've been able to correct the situation a little. We managed to get read speeds to the same decade than what 3ware gives out. Write speeds are still a little too slow, but much much better. No major source of problem was identified. (I actually do not know the details anymore... this was out of my leaque :-) but read-ahead settings needed to be fixed in all the right places. But still the loads seem to be up. The situation now seems to be so that if there are lot's of processes, then the processors do not spend time in i/O waits, but if there are not that many processes and processors can idle, then they will still spen uptp 80% in I/O wait (not idle). The setup we have is - 8 disks in SATA-RAID visible to the SuSE as one large Logical Disk. - /boot is on it's own partition on the RAID - everything else is configured as one LVM - on that LVM we have separate partitions for / (root), /swap, /var and /loca (almost all of the rest) Is there something fundamentally wrong with this? Or should this kinf of setup work OK? -- Hugo.