On Thu, Jan 5, 2012 at 4:11 AM, Henry Dogger <[email protected]> wrote:
> Is this normal behavior? A server under this kind of load will eventually
> start using its swap memory and then failing (we tested this).

I saw the same behavior.  We're talking about RSS or resident memory I
assume.  Virtual will always go up.  Joegen found a couple places
where memory was leaked and there were several updates to 4.4 to
address this.  Are you running the absolute latest 4.4 update?

Overall, there were many places where large amounts of memory were
used because entire copies of the database (FreeDB) were made into
memory.  In 4.6 we now leverage the searching database capability of
the new database (MongoDB) which means we only have to load what we
need. We should see leaks go away and an overall decrease in memory
usage of 4.6 v.s. 4.4.  It should be quite dramatic especially with
systems with many users (~20K users).

I won't even get into the fact that supervisor was using an obscene
amount of threads (2x more than any other process) and questionable
amount of memory.  Again no leaks, just the nature of the architecture
choices made. in 4.6 the memory will decrease from GB to K. and
threads will go from around 170 to 1.

Just to finish the topic, we saw occasional saw deadlocks in 4.4 with
no smoking guns.  Again, because of the underlying db change to mongo
which leverages threadsafe connection pool that comes with mongo api
and the fact we have less shared memory, nearly every single lock
could be removed in 4.6, at least the ones in the path or message
processing.

In addition, in 4.6 you'll be able to turn off a lot more services
that you do not need.

So hang in there...

-----------------
Join me to talk about sipXecs and the upcoming version 4.6 at CoLab @ CSU in
March (5th & 6th).  http://www.sipfoundry.org/sipx-colab
_______________________________________________
sipx-users mailing list
[email protected]
List Archive: http://list.sipfoundry.org/archive/sipx-users/

Reply via email to