topic Re: Slow Performance in Operating System - HP-UX

Slow Performance

Jade Bulante — Fri, 08 Dec 2000 05:01:03 GMT

I have noticed that whenever a report is being runned on our HP9000 11.00, all the users complain that the system slowes down. All the users are hitting the same database on the same disk and I have a feeling that it's most probably a disk performance issue. I re-verified the bottleneck using sam and I saw the bottleneck on the disk. Is there quick fix to this? Can I modify the kernel?? Can I add even out the disk bottleneck??

Re: Slow Performance

+¦+ó+± — Fri, 08 Dec 2000 05:52:31 GMT

Hi,
If oracle database, you can check heavy IO's dbf file. In this case, you'd betther split dbf file to other disk.
Here's my some considerations.
- enlarge buffer cache size
nomally, dbc_max_pct : 20, dbc_min_pct : 5
bufpages : 0
- fs_async : 1
- VxFS mount option
- IO channel split
- disk striping
etc.

Re: Slow Performance

Dan Hetzel — Fri, 08 Dec 2000 06:18:00 GMT

Hi Jade,

It's not surprising if you have all your database files on the same physical disk.

Unfortunately, the only way to solve this issue would be to spread your database onto multiple disks, relocating datafiles, control files, redo logs... trying to distribute them evenly.

You'll need the help of your DBA to achieve this as it's nos as simple as moving the files.

Best regards,

Dan

Re: Slow Performance

Michael McCauley — Fri, 08 Dec 2000 12:55:26 GMT

You also might want to consider stripping across multiple spindles and multiple controllers if you have them available

Re: Slow Performance

Dave Wherry — Fri, 08 Dec 2000 15:20:28 GMT

It may not be just your hardware configuration. If your database is on one disk you certainly have an area for contention.
However, how efficient is the program? Is it using indexes? Sequential scans and "select * from..." absolutlely kill performance.

Re: Slow Performance

CHRIS_ANORUO — Fri, 08 Dec 2000 16:22:54 GMT

Take a look at these documents: S3100002312A, B and C. You can run the attached script at the peak periods to determine the actual cause of the bottleneck. Modify the kernel as specied on this thread including nbuf and bufpages=0, and set swapmem_on value to 1(one).
Glance will give you good insight into the problem on the ground.