[mrtg] Re: Solaris Related Problems Off topic post - sorry

[DApel at omnipoint.com]

For accurate information about where Solaris is using memory, try Richard
McDougal's RMCmem package (aka MemTool).
 
His email is Richard.McDougal at eng.sun.com
<mailto:Richard.McDougal at eng.sun.com> 
 
If I remember correctly, you have to email Richard to get the package,
because Sun doesn't support it.
 
I was tipped off to this by one of Adrian Cockcroft's articles regarding
Solaris performance tuning.  http://www.sun.com/980922/tuning2/about.html
<http://www.sun.com/980922/tuning2/about.html> 
 
He recommended Richard's program, and I must say it's really useful.   I use
it as in my Big Brother stup as a client-side test.
 
As far as pulling this data via SNMP, I'm afraid you'll have to get
creative....
 
Here's some sample output of an Ultra 2, Solaris 2.6, with 560MB of ram
(512+32+16).
glass # /opt/RMCmem/bin/prtmem
 
Total memory:             557 Megabytes
Kernel Memory:             35 Megabytes
Application memory:        69 Megabytes
Buffercache memory:       242 Megabytes
Free memory:              211 Megabytes
 
At the very least, I can show myself AND my users that they really don't
need additional memory, and I don't have to play "adding machine" totalling
up memory used per process.....
 
BTW, Solaris (SysV-ish) handles memory "better" than some other OS's such as
HPUX which are more BSD-ish.... Under Solaris, you do not need to have swap
be two times physical ram, which of course has been the old rule of thumb.  
The next time you install a 512MB ram box for instance, if you manually add
a swap partition to the disk tables, you'll notice that the "recommended"
size might only be 32MB, or 128MB, or some such number far less than the
physical ram.   I like this because I don't have to blow a whole disk in
order to have it sit idle.   Disks are cheap, but they're not free!
 
For instance, HPUX performance takes a huge nose-dive if swap space isn't at
least twice as large as physical ram.   This is apparently due to a
kernel-level allocation routine.   If you don't have the 2x+ swap, the CPU
will (according to HP ESC) waste a good percentage of it's time TRYING to
allocate swap space, even if it never gets used.   I've got a K420 server
with 768MB of ram, and used to have a 1024MB swap partition (the box used to
have 512MB ram).   Performance after adding the extra 256MB of ram was
actually worse than with just 512.   No swapping actually takes place
according to swapinfo.   However, HP finally asked me how big the swap file
was -- I increased it to 2GB, and performance went up by 38%!!!!!   Of
course, the down side is that I have 2GB of disk that is just WASTED because
the server never actually swaps more than about 100MB, and that's rare.  
Because of this, I have a mountain of old seagate 2gb FNSE disks to use for
swap disks -- if they fail, I've got lots of spares, and they're too small
and slow to be really useful anywhere else.
 

Doug Apel 
Sr. Network Administrator 
Omnipoint Technologies, Inc. 
dapel at omnipoint.com 

 

-----Original Message-----
From: Charles Gillanders [mailto:charles at toucan.ie]
Sent: Wednesday, July 28, 1999 10:01 AM
To: MRTG Mailing List (E-mail)
Subject: [mrtg] Solaris Related Problems Off topic post - sorry



Hi this is an off topic post, sorry but I have no idea where else to ask. 

I'm successfully using MRTG and the UCD snmp libs  monitor disk space,
processor usage and free real and virtual memory on a solaris 2.6 system.
MRTG works great - thanks Tobias for a great little tool.

It's what mrtg shows up that has me worried, the real memory free very
quickly falls from near 500Mb (the box has 512Mb in it) to just about 7 - 8
Mb.  If I run top on the system and sort everything by size rather than cpu,
I can count up all the memory that the various processess are using, guess
what it doesn't add up to anywhere near 500Mb, more like a 100 or so.

So my question(s) is(are) where has my memory gone and should I be worried? 
Does anyone else monitor similar stats with a sun and notice similar
results?  Does anyone have any pointers for mailing lists / web sites /
usenet groups etc. that might be able to advise?

I should point out that having seen the graphs and not believed them I
wondered if the UCD libs were giving me bum data, I ran top at intervals
over the 30 minute period or so following a reboot, and top also shows this
steady decrease in the amount of free real memory available.

Thanks, 

Charles 

--
* To unsubscribe from the mrtg mailing list, send a message with the
  subject: unsubscribe to mrtg-request at list.ee.ethz.ch
* The mailing list archive is at http://www.ee.ethz.ch/~slist/mrtg