Original post:
What are the criteria for determining if a system is seriously
short of memory? When I run sar -p I see short bursts of
activity in which vflt/s value goes up to around 140, then drops to
0-10. The pflt/s statistic tends to run from 0 - 40, following approximately
the same pattern. Is this high? Are there other statistics that I should
monitor?
*************************************************************************
Summary:
Thanks to everyone who responded so quickly.
Almost everyone who replied suggested the SE toolkit. I got it from
http://www.sun.com/960301/columns/adrian/column7.html
It is very easy to set up and use and provides extermely usefull information.
The tool that I particularly like is zoom.se. You start it up and
leave it iconized. If any of the parameters that it is monitoring
(disk, memory, network ... ) reach danger levels, the icon changes
to a symbol for that resource and starts flashing. Since I am
monitoring five remote systems, this is very usefull.
My analysis so far indicates that the performance problems had no
relation to memory, cpu, or disk. It was all Oracle. Although this
is somewhat off topic, I will give a brief summary of what I found and
what I did to cure (I hope) the problem since it may help others.
(Begin off topic information )
Our application runs on three workstations accessing some tables
on the local node and some remotely via database links and synonyms,
I had my databases configured as multithreaded servers, with
the mts_servers parameter set to 1. This meant that the Oracle dispatcher
was constantly having to create and destroy additional servers to handle
the workload from multiple processes. I could observe this using
Oracle's server manager - Monitor Servers window. I increased the
mts_servers parameter to 6 which was the maximum number of servers
running. Performance problem disappeared.
If anyone wants more details on this I can provide more details off line.
(End off topic information )
*************************************************************************
Details of responses:
********************************************************************
From: Alex Finkel <afinkel@pfn.com>
Subject: Re: How can I determine if a system is memory bound?
X-Lines: 33
Status: RO
If you are running Solaris 2.x, get a hold of the SE Toolkit. There are
some example utilities that will monitor several stats on the system and
indicate where help is needed.
The following URL has links to download the latest SE and to information
about it.
http://www.sun.com/960301/columns/adrian/column7.html
- Alex
.
********************************************************************
From: John Benjamins <johnb@Soliton.COM>
the system is constantly paging, and therefore is getting page faults
(that's how it knows it needs to load in another page). how many page
faults you get tells you, basically how active the system is, and this
will not really be affected by the amount of memory you have, unless
you are very, very short on memory, and the system is performing
desperation swapping. however, i don't think that the sunos kernel in
solaris 2.4 or 2.5 will ever swap.
run "vmstat 10" and monitor the sr column. if this is consistently
over 200, your system is very short on memory. this comes from the
sun performance tuning book by adrian cockcroft. if you are doing
performance tuning on a sun/sparc, get the book. it's very
informative.
********************************************************************
From: zaitcev@lab.sun.mcst.ru (Pete A. Zaitcev)
sar is overkill, use vmstat -S for begining. Watch out for unnecessary
pageout. Scan rate also gives a clue.
********************************************************************
From: "Daniel J Blander - Sr. Systems Engineer for ACS" <Daniel.Blander@ACSacs.Com>
Virtual Adrain (SE Toolkit) which is available from the Sun
Web Page is a good judge of RAM, Swap and all the settings.
It will even let you know if system/kernel settings will
really fix things....since many of these rules take years
to understand and I am no such expert, you may try Addrians
tools (I did and its pretty cool). The best basic tool out
of the tool kit is to run the script zoom.se (just like that at
the command prompt)....
********************************************************************
From: Kevin.Sheehan@uniq.com.au (Kevin Sheehan {Consulting Poster Child})
Yes - get Adrian Cockcroft's "Solaris Tuning and Performance" book.
The single most important field is the "sr" (scan rate) field in
vmstat/sar. Faults occur all the time, and are just the way business
is done - scanning is what the page daemon does when the system is short
of memory. More scanning means more memory shortage.
********************************************************************
From: Ric Anderson <ric@rtd.com>
For a decent real time picture, get and install "top" from
ftp://eecs.nwu.edu/pub/top/top-3.3.tar.gz
it will show you real mem usage in the first few lines of the
display, as well as time spent swapping.
********************************************************************
From: Asim Zuberi <asim@psa.pencom.com>
Hi!!
What OS you have. If you have Solaris I have a utility called SEtools
which I can mail it to you. This is a product from Sun to monitor RAM,
network, NFS etc.
********************************************************************
Peter Schauss
ps4330@okc01.rb.jccbi.gov
Gull Electronic Systems Division
Parker Hannifin Corporation
Smithtown, NY
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:11:09 CDT