topic Re: swapper constantly at 4-5 % usage in Operating System - HP-UX

swapper constantly at 4-5 % usage

Youlette Etienne_2 — Tue, 31 May 2005 11:40:51 GMT

This seems a bit strange. The server has 16 gb or memory, and currently has over 8gb free. There are 2 Oracle databases, but this is a test server and there is very little if any user activity.
Swap space is at 5% useage, with 30 gb available. vmstat is showing no paging occurring.

kmtune|grep dbc
dbc_max_pct 20 - 20
dbc_min_pct 5 - 5

Load averages: 0.15, 0.14, 0.13
190 processes: 166 sleeping, 23 running, 1 zombie
Cpu states:
CPU LOAD USER NICE SYS IDLE BLOCK SWAIT INTR SSYS

avg 0.15 0.0% 0.0% 1.0% 99.0% 0.0% 0.0% 0.0% 0.0%

Memory: 1603148K (951272K) real, 2215960K (1495452K) virtual, 8995956K free Page# 1/15

CPU TTY PID USERNAME PRI NI SIZE RES STATE TIME %WCPU %CPU COMMAND
2 ? 0 root 127 20 32K 0K sleep 2660:20 4.25 4.24 swapper
1 ? 3 root 128 20 32K 32K sleep 5075:17 3.83 3.82 statdaemon

I have another server with half the resources and 11 oracle databases installed. Swapper is around 1%, with swap space at 42% used on this server.
Is this strange?

Re: swapper constantly at 4-5 % usage

RAC_1 — Tue, 31 May 2005 11:48:28 GMT

I would advise not to worry about it. Every processes that starts will reserve some swap space. This is required because, in case it needs to swap out, it will have required swap space.

Do you get any swap outs (a number that you need to worry about. e.g. say 10 or so)??

What does swapinfo -tam looks like??

Anil

Re: swapper constantly at 4-5 % usage

Jeff Schussele — Tue, 31 May 2005 11:58:02 GMT

Hi,

Is it possible that these Oracle instances - or any other applications on the system - are using MRGs (Memory Resource Groups)?
If so they have their own physmem_pages and subsequent minfree/desfree/lotsfree values & the sched() (known as swapper) process would have to handle it's memory management just as if it was it's own system.
See the Memory Management Whitepaper for a detailed explanation - here:

http://docs.hp.com/en/1218/mem_mgt.html

Rgds,
Jeff

Re: swapper constantly at 4-5 % usage

Steven E. Protter — Tue, 31 May 2005 12:02:38 GMT

You might be able to get that swap use down with one bold stroke.

dbc_max_pct 20 - 20

Change it to 10 or even 7 or 8.

That will drop memory use and reduce the amount of double buffering with oracle's cache.

It will also reduce the size and serverity of changes in the cpu expensive process of changing the size of this cache.

SEP

Re: swapper constantly at 4-5 % usage

Patrick Wallek — Tue, 31 May 2005 12:14:18 GMT

Everyone so far has misunderstood the post, I think.

The complaint is that the swapper process is using 4% t0 5% CPU on this machine constantly.

That does seem high to me. You might check for any applicable patches for the swapper process.

Re: swapper constantly at 4-5 % usage

Jeff Schussele — Tue, 31 May 2005 12:34:27 GMT

Hi Patrick,

I understood the question.
If the user had processes (or groups of processes) like a specific Oracle SID or even another application on this server set up to use MRGs then swapper would have to handle it just as it handles a "normal" system.
And quotung from the whitepaper

With Memory Resource Groups, vhand and sched behave almost as if each MRG were completely separate, with its own individual pager and swapper. (The actual implementation is a bit more complex, as it must account for processes and memory moving between MRGs, the ability for one MRG to borrow memory from another, memory use that can't be assigned to any single process (or any MRG), and the need to maintain global memory availability as well as individual MRG memory availability.)

\Endquote

Well then old swapper could be a very busy beaver. And that could certainly explain why he is w/o any "actual" paging out. Although you'll note that swap usage is at 5% so one has to assume that there is at least *some* paging going on.

My 2 cents,
Jeff

Re: swapper constantly at 4-5 % usage

Youlette Etienne_2 — Tue, 31 May 2005 13:57:31 GMT

I will check for relevant patches. I do not see how this can be an Oracle memory management issue that can affect the swapper process as suggested. The 2 databases are up, but they are not being accessed at this point in time because there is currently no data migrated to it yet. I would find it very strange that a database is extremely busy doing nothing. If it is, then I need to consult witht the DBAs.

There are no other applications installed on the server.

Swap space is not the problem. 5% usage seems quite reasonable for a system that is not doing anything but running the Oracle processes. Most of the answers have misunderstood the question.