You don't mention what kind of session affinity you have configured... I only mention this because your key problem seems to be with session affinity - and pound offers several types...

 

Personally I would think that a session type of "IP" (because a hashed lookup should be the fastest) would be a reasonable compromise for high performance against equal balance in order handle peak loads the best on a modest server.

 

The drawback is "if" you have lot of users from a single IP, then your cluster will not be 'perfectly' balanced. i.e. some backend servers will have to handle more load than others.

 

We have 10 times your traffic _average_ by the way on a _very_ modest (single CPU) server and no problems whatsover :-)

 

/David

 

On 10/28/05, Russell Valentine <russ@altec.org> wrote:

Hi, I've been using pound for about two years, thanks for the great
program. I have a question concerning sessions. I've been noticing
during our busiest times it seems after a certain point pound can't
remember new sessions anymore, so new sessions cannot be maintained on
the web server.

1) User hits pound
2) pound sends request to certain backend
3) session gets made on backend
4) user hits pound again
5) pound sends request to a different backend, now backend has no idea
about the session

I only see this problem during times when we have most of our traffic,
when we get around 1000 requests per minute.

I tried to see if we were perhaps maxing on threads or file descriptors.
I counted the treads with ps and used lsof for file descriptor and
counted those periodically in a cron job.

Max I see from lsof: ~200
Max # of threads I see is ~100

These are no where near the limit as shown in ulimit.
I'm using pound v1.9 on Linux 2.4 kernel. Does anyone have ideas?


Russell Valentine

--
To unsubscribe send an email with subject 'unsubscribe' to pound@apsis.ch.
Please contact roseg@apsis.ch for questions.
http://192.168.1.2:8080/Apsis/pound/pound_list/manage_mailboxer