Performance I/O disk problem

[Juan_78]

Hello,

I have a HP server rp5430 serie HP9000 running OS HP-UX 11i, with 1GB RAM, 2 CPUs and a disk controller RAID4si.

This server has installed an e-mail service for 50 thousand accounts (software is qmail + vpopmail + mysql + Maildir).

At this moment, we are experiencing a problem of slowness in the reading and writing of disk,

Apparently because is much the information that receives (e-mail service). Is generate a bottle neck in the writing to the disk, for that reason my e-mail service is slow for local messges.

When I use the glance command, I saw that the disk used arrives at the 100% and memory 99%.

I need to improve the performance, making an adjustment or tuning kernel, adjustment of the disk, or another option.

Previously I don't have any problem with another equipment, this was an Alpha 1200 Server with 739MB RAM.

I don't know if I should change some kernel parameters, for example: fs_async, nfile, maxuser, nproc, dbc_max_pct, bufpages, etc.

I attachment some additional information that hope serves to you for the diagnosis.
I wait for your recommendations.

Very Thanks

[Michael Tully]

The first thing you need to do is reduce the size of your buffer cache to around 300Mb. (dbc_max_pct) This will have an immediate effect on the memory. You could start the setting at 30 for dbc_max_pct and say 15 for dbc_min_pct. 3 out of 5 disks are getting thrashed, which indicates a balancing problem. Perhaps you could identify your volume group information. I don't think changing the mounting options for filesystems is going to make a deal of difference. It could be the controller getting trashed or you may need additional disks to spread the load. What application is being used ?

[Stefan Farrelly]

Mail is completely i/o bound, which means you really need to setup your disks to be striped at the Lvol level across as many disks as possible and across as many disk controllers as possible.

Are the HP disks as fast and with as much cache as the ones on your Alpha were ?

What is your disk layout like ?

[Juan_78]

Thanks for your response
Michael and Stefan

My system have this configuration:
one controller disk RAID4Si with 4 disk of 36GB in raid5

Michael: The application used is qmail, this application have hundreds process, and the problem appears when arrives the local e-mails. In this moments the local delivery is slow, and disk used is 100% and %busy too is 100%.

Stefan: I am attaching my disk layout (command vgdisplay -v vg01)

[GK_5]

You need to upgrade memory on the box. 1GB is very less for what is running on the box.
Also change the swap accordingly.

To reduce disk I/O use striped mirror or RAID5.

[Stuart Abramson_2]

You are I/O bound on disk c4t0d0.

1. Rebuild kernel with max_dbc_pct set to 10%, as previously mentioned. This is NOT the prime culprit, but should be done.

2. the "sar -d" shows that your system is beating up on disk c4t0d0, but c1t0d0 isn't much used. SPREAD YOUR DISK I/O ON THE OTHER DISKS.

If possible, spread your disk i/o on another controller. Multiple paths. Stripe across multiple paths if you can.

The sar command only shows what's happening now. Get an "over time" look at your i/o and try to spot the hot disks....