Glance Plus easy question 3

[ehspacl]

If Mem Util is at 99% is this bad and can I do anything to lower it, maybe raising the Swap Util some how?
Swap Util is at 50%.
Disk Util is at 92%

Thx

 

[dobbyn]

Most likely your performance is suffering and most likely you need more memory for whatever applications you are running. Swap space is used by the system when processes cannot all be kept in memory, so increasing swap will not free up any more memory. The high disk usage number probably indicates your system is already spending a lot of time swapping - this is very costly.

Depending on the applications you are running (Oracle, for example), you may be able to adjust how much physical memory it uses, but probably at the cost of performance of that app. You might also look at the kernel parameters for buffer cache (dbc_max_pct, dbc_min_pct, bufpages, and nbuf) to see if you are allowing the system to use large amounts of memory for file cache.

 

[ehspacl]

Many thanks dobbyn

 

[bpk89]

You say GlancePlus is displaying a total of 99% utilisation. This can mean that you are seriously out of RAM or getting good value for money in terms of the amount of RAM purchased.

Glance breaks up the memory utilisation bar into three parts. (i) S - System mem (ii) U - User mem and (iii) B - buffer cache. All of these combined give you total mem utilisation.

If your buffer cache is set dynamically with no upper limit then it may consume all available RAM and display 99% utilisation. The buffer cache will hopefully (but not always) hand it back when the system or user mem needs it.

What is your buffer cache set to? Check the kernel parameters as mentioned above in dobbyn's reply.

Next. Check for swap activity. To me 50% doesn't seem too bad. Look at Glance (text version) press "w" to see swap. A better approach would be to check the memory report F3 or "m". Look for activity on the "KB Paged Out" row. Typically any paging out is bad.

Next. You say you have 95% disk utilisation. Do you have multiple disks. Check which physical disk is doing all the work. You may find that one disk is flat out and the other is virtually idle. Use "u" I/O by Disk and check it out. Can you move your logical volumes over to different disks and spread the load?

Last. You didn't mention CPU was high therefore I'd make an assumption that CPU is not processing swap and is coping quite well with the load its given.

My conclusion is that your performance problem is caused primarily by a disk bottleneck and not a memory bottleneck.

 

[ehspacl]

Thanks bpk89,
The info. you asked for is:-
dbc_max_pct 50
dbc_min_pct 5
bufpages 0
nbuf 0

swap usage:- Available Used Priority
/dev/vg00/lvol2 1GB 0 1
/dev/vg00/swap2 2GB 354mb 0
pseudo-swap 1.1GB 950mb NA

Memory report:- Cumulative High Rate
KB Paged OUT 80kb 0.9

The HP-UX box has a total of 1.5GB over two CPU's
Clock Frequency is 440Mhz

I have got five instances of oracle 8i installed over six 18GB physical disks. Each instance is separated over one or two physical disks.
My shmmax is set to 1GB and shmmni is set to 200.

I hope all this info can help you diagnose the problem.

I really appreciate your help.

Many thanks in advance.

Paul

 

[johnknapp]

I just went through some kernel tuning with HP Technical Support and the overwelming concensus is the Buffer Cache setting should not go beyond 350MB to 450MB ... going beyond that amount wastes memory for Buffer Cache unless your machine is solely an NFS server. I had a similar situation in which over 1.5GBs of RAM was dedicated to Buffer Cache (dbc_max_pct) out of 3.5GB of RAM because HP/UX default setting for dbc_max_pct is 50%. My memory usage showed up well over 90% utilized.

I ended up setting dbc_max_pct down to 10% (350MBs). HP strongly suggests setting buffer cache at that level unless you have smaller amount of RAM (512MBs).