> Now, one of my questions remains a mystery: why did
> the system initially stop 
> giving out virtual memory when it hit approx half of
> its RAM (8Gb of 16Gb)
> and no swap space was enabled? Was it just a
> coincidence?
> 
> Jim Laurent suggests that Solaris can be configured
> without swap whatsoever,
> and I thought so too - until I hit these problems
> above ;)

Been there, done that.

I bet zfs complicates things, given that it will suck up quite a lot of memory
(although it's supposed to give some back if things get too tight).

> Is there some tool to check which processes reserved
> virtual memory and how
> much? I played around with "ipcs", but I guess I
> didn't find the correct options 
> yet... or this should be seeked elsewhere. Or is this
> exactly what the SWAP
> fields in "top" and "prstat" are about? :)

Since the non-bundled RMCmem (memtool) was last updated only
for something like Solaris 9, most other tools are less pretty.  pmap
can report on individual processes.  prstat or top or vmstat can report
on the system as a whole.  prstat or top or ps can also give info about each
process's usage.  The problem is understanding all those - they're not
necessarily measuring the same thing nor in the same units, so it gets
confusing.

> Also, another question arose. I found that /tmp
> filesystem size can be limited 
> by settings in /etc/system, i.e. allowing /tmp to
> consume no more than 200Mb.
> Is it possible to instead "reserve" some virtual
> memory for filesystems like /tmp,
> /var/run, /etc/svc/volatile and so on - so they would
> have some small memory
> to work with even when the processes have not enough
> virtual memory to fork?
> 
> Many of the services seem to die, or can't restart
> after crashing, because they 
> can't update some lock file in /var/run or for
> similar (FS-based) reasons :(
> 

AFAIK, there is no mechanism for guaranteeing availability of VM for tmpfs,
only a mount option limiting how much the tmpfs resulting from that mount
may use.  Until/unless there were, the only way to guarantee such a thing is
to place limits on everything else, which may not be realistic.  One can always
just avoid using tmpfs (although what to use instead goes beyond where I want
to go in this post).
-- 
This message posted from opensolaris.org
_______________________________________________
opensolaris-discuss mailing list
[email protected]

Reply via email to