System slow

[tyagineeraj]

Hi,

HP K-400 production sever becomes slow and sar -v is giving me inode-sz 2000/2000 or 1998/2000 most of the time. and sar -u is showing %wio to be 40-60 when system becomes slow.also sar -d for the same time gives two RAID 5 disks groups busy 30-50%. DO you think cpu is waiting for disk i/o. if so can i break RAID 5 TO RAID 0+1.

 

[ianf]

Do you really think that moving from RAID5 to RAID 0+1 will speed up disk I/O ? Surely you're moving from single read/write using multiple disk caches in RAID5 to double r/w on individual disk caches in RAID0+1? That's unless I'm missing something.

 

[tyagineeraj]

Actually Raid 5 calculates parity so writing becomes slow.But i do not know how to break Raid 5 to 0+1 on HP as Raid 5 is not supported by LVM it is hardware raid. But my first concern is if machine is slow because of disk i/o or some kernel parameters can be adjusted to improve performance.

I sad it seems to be disk i/o because wio time is high and i do not find much swap activity.

 

[ianf]

I'll bail right there then !

 

[MikeLacey]

I think you're on the right track there tyagineeraj, and you should definitely, after some careful thinking and research, consider increasing the size of your inode table kernel parameter.

Questions though:

1. Your h/w r5 thingy -- will it support r0+1?

2. How much memory do you have? How much of this is dedicated to filesystem cache? (or is that allocated dynamically, the default in fact)

3. What applications are running on your server?

4. How many concurrent users? Mike
michael.j.lacey@ntlworld.com
Email welcome if you're in a hurry or something -- but post in tek-tips as well please, and I will post my reply here as well.

 

[tyagineeraj]

Hi Mike,

Here is some information for you and some other things:

1: still to find if it support 0+1 h/w but right now i do not have enough space to break r5 to 0+1. So i am trying to look options for improving disk i/o.

2: memory is 532mb. How can i see how much is dedicated for file system cache.

3: Filenet is running with oracle 8.0.4.
4: 200 to 240 users mostly of filenet.

I noticed kernel parameters fs_async = 0, create_fastlinks
= 0. Do you think chainging fs_async = 1 so puting write of file system meta structure into asynchronous mode will improve i/o. System is connected to ups. what about create_fastlinks?

Thanks For reply
Neeraj Tyagi

 

[rjdscs]

If you have the HPUX disk set there should be at least one of the CDs with demo software. On one of these CDs you should find a pakcage called glance+ which provides a lot of performance data. It will run as a demo for about 60 days, or there abouts. You can purchase a license for it from HP. It will show memory, swap, cpu, network performance stats. It will show how much memory is being used by the kernel, how mush by processes, how much cache, etc. It has botha graphical and text interfaces. The graphical interface provides a considerable amount of data and graphs.

Richard

richard@brecon.co.uk