performance issues!!

[7280]

Hi,
i have a RISC/6000 with aix 4.3.3, 1024 MB RAM, 2048 swap, two controllers which connect to ssa disks.
From topas i see this machine has always disks busy between 30 and 90%. Memory is always 25% comp, 75% Noncomp. Paging 60% used, 40% free. But in paging i see somethig like this:
Faults 500
Steals 1000
PageIn 700
Sios 570
And of course cpu is always idle about 0.5 %.
i have about 60% user cpu, 30% wait and other kernel...
So i'd like to understand what's the problem.
I think it's memory because it's always paging. But then i see i have 75% memory free.
Are the disks too slow?
Please advice me.
Thanks

 

[bi]

Are all the disks 60-90% busy or just some of them? Where do you have your paging -- on internal disks or the SSAs? What kind of application is on the SSAs?

 

[7280]

most of ssa disks are busy at that percentage.
internal disks are busy about 20-30%
on internal disks i have operation system and swap.
on ssa disks there's a database application.
what i'd like to know is if it's only i/o problem or also memory.
thanks a lot.

 

[bi]

It might be a little of both. I'm assuming you are asking this because of complaints about performance?

The busy internal disks are probably caused by the paging. If you had more memory you might not have as much paging. 1 GB of memory isn't that much these days.

You might try tweaking some vmtune settings to help improve performance.

It might be an i/o problem because the disks are so busy. If the database files are not distributed across the disks in an optimized way (keeping much used tables on separate disks, and rollbacks and redos on separate disks and away from other write intensive files, for example), you will see a performance hit. If write intensive files (like rollbacks, redos, and data files) are on RAID5, you will see a performance hit.

I also would be concerned about the high percentage of wait on the CPUs. With 30% wait, you've got lots of processes that are waiting to get to a CPU to do what they need to do. When I have had systems get to near 0 idle for the CPU, the system is just crawling.

How many CPUs do you have, what model is this RS6000 and how old is it?

 

[bi]

Also, what is the speed of the CPUs and do you know the model of the CPUs (something like RSIII, for example).

 

[7280]

I don't think it's CPU problem. I think CPU's wait time is high because of "the problem". For "the problem" i mean that maybe it's waiting for free memory, mabye it needs to write but the disks are already busy, maybe it needs to read but disks are locked for writing.
I don't know... I think the CPU will be a problem if USER % is very high, so it's using all available CPU.

 

[dblair]

What type of databas ?
How may instances ?
Size to database ?
How many disks (logical) avaiable to O/S.
Raid level ?

 

[7280]

It's a banking application based on MUMPS database.
3 instances running, two instances of 12 gb and one of 1 gb.
There are two internal disks and 4 ssa disks.

 

[dblair]

Are the 3 databases on the SSA disks. ?
What Raid level are these disks ?

 

[murby56717]

7280
I was curious if you could include a vmstat and iostat output. It might help the group narrow down exactly what is happening.
Best of luck
Bobby