I was never particularly happy describing this as a
"leak", since the memory gets freed up again. Turns
out that I would have been right not to, since this
has never been a leak.
Thanks to John Leadeham, Pete Durst, Robert Alexander,
Buddy Lumpkin, David Spence and Dennis Martens for
their replies. I think Pete Durst's reply explained
this best, and I'm just going to quote it directly:
> You are not seeing a memory leak, but in fact seeing
> how memory is being used by the system. Your
> observations were right on. The systems maintain
> a minimum number of pages of memory in a free memory
> pool. While the system has more memory available
> than this level, it does nothing to replenish the
> pool. Once the level falls below the minimum
> threshold, it goes out and gets the un-used memory
> and puts it back into the pool.
> The threshold is set to 1/64th of total memory, and
> this will explain the differences between systems.
> The reality of the system is this: You can never
> really, truly know how much memory is free, as the
> system only looks for free memory when required
> (less than the threshold) and stops looking once it
> has more than the threshold. So, in most cases, you
> will see the system hover around the value of the
> threshold. (The value you see is the number of
> pages available, and needs to be multiplied by
> `pagesize` to see the actual size of the free memory
> pool). Adrian Cockcroft's Sun and Performance
> Tuning book is an excellent resource to have close
> at hand.
>I've seen this issue on a number of 2.6 machines, and
>am trying to find out where the problem is coming
>While gathering information for system
>troubleshooting I will often run simple resource
>In this case:
> while true
> date >> vmstat.out
> vmstat >> vmstat.out
> echo >> vmstat.out
> sleep 900
>What I've seen on a number of 2.6 systems is that
>used virtual memory will steadily increase to a
>certain threshold - different values on different
>systems - and then drop back to zero(ish) again.
>This doesen't seem to have any impact, but it happens
>on systems at various patchlevels, I can't find
>anything on Sunsolve (though I'm maybe looking for
>the wrong thing) and there don't seem to be any
>processes growing apace. I'd just like to find out
>if someone knows where this "leak" originates from.
Do You Yahoo!?
Talk to your friends online and get email alerts with Yahoo! Messenger.
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:14:06 CDT