SAMBA eats up all memory...
jojowil at hvcc.edu
Thu Jul 6 12:37:50 GMT 2000
Sumitro Chowdhury wrote:
> 1. (A) high water mark and low water mark are set to 0
> (B) maxrandwrit = 0 in vmtune
> So I would say I/O pacing is off and write behind is off.
Okay...I thought so...
> 2. Since avm in vmstat is not increasing, there does not seem to be
> any memory leak. but the system is CERTAINLY running out of memory.
> This is what is frustating that I canNOT "see" how system can run
> out of memory without a) memory leak b)heavy paging.
Actually, that's not entirely accurate. This is the amount of memory in use by
programs, not AIX proper. IOW, if you have:
perfagent.tools 126.96.36.199 COMMITTED Local Performance Analysis &
installed, run the following (if not, I *strongly* suggest you get it from the
[storage:/] # svmon
size inuse free pin virtual
memory 784359 742339 29687 784359 63734
pg space 786432 32823
work pers clnt
pin 30423 0 0
in use 64830 677475 34
[storage:/] # vmstat 2
kthr memory page faults cpu
----- ----------- ------------------------ ------------ -----------
r b avm fre re pi po fr sr cy in sy cs us sy id wa
0 0 63479 29957 0 0 0 10 22 0 134 380 66 1 2 93 4
0 2 63479 29955 0 0 0 0 0 0 435 2412 55 0 1 98 0
0 2 63479 29955 0 0 0 0 0 0 431 211 49 0 0 99 0
0 2 63479 29955 0 0 0 0 0 0 444 215 53 0 0 99 0
As you can see, I have a 3GB system, but only 250+MB is tied to processes
(work), the rest is in file caching (pers) or is free.
> 3. lsps -a:
> Page Space Physical Volume Volume Group Size %Used Active Auto
> paging00 hdisk1 rootvg 1024MB 1 yes yes
> hd6 hdisk0 rootvg 512MB 1 yes yes
> Essentialy there is no disk paging.
Which makes sense, since maxperm is 10% - the VMM will leave working pages alone
and aggresively steal file pages to minimize paging. What you did was correct
but overkill. I would leave maxperm at 80%. This would give 6.4GB to files and
1.6 to programs and the kernel. The system should still aggresively steal file
pages and not do swapping since you only require ~1GB at present. If paging does
begin, reduce maxperm by 5% until it stops and levels off but try to stay at 50%
or higher since the VMM is designed to page to disk as well and is pretty smart-
you should always look at avm from vmstat to see exactly what your processes
> 4. vmtune output:
see above ;)
> 5. I would also tend to agree that MACs are screwing things up but they
> require proof. What troubles me is that when MACs and NT stop
> writing to the shared file system, I can't copy , move , rm 50MB
> files from the AIX prompt even. I get system out of memory errors on the
> screen. Nothing on errpt though and as u saw, avm is around 1GB.
> What's happening to the rest 7GB of memory ???
Check out this web page from IBM. It'll help you discover memory leaks in
Use smbstatus to get the pid's of suspicious clients and get the proof you need.
Check out the online docs here:
There is more:
What's the current number of processes running and what's the maxuproc value of
lsattr -El sys0? Are all your smbd's running as root (which they should be)?
I ask because this is an argument I have with IBM right now that maxuproc as of
4.3.3 (but not 4.3.2 or lower) seems to affect root (uid 0) when it should not.
Also what are the ulimit -a values for root and everyone else? these are also
stored in /etc/security/limits. You may simply be hitting a data segment or rss
wall which would look like your system is out of memory or, more to the point,
like you have a memory leaky program (which I really don't think you do...the
Samba Team has worked their asses off to make sure this code is clean and fast)
More information about the samba