[imp] session problems & memcached

[Kevin Konowalec]

Horde: 3.1.2
Imp: H3 (4.1.2)
Turba: H3 (2.1.1)
PHP Version: 4.4.1

We've been getting complaints over the last few days of more and more  
people being booted out of Imp frequently.  We haven't set any sort  
of timeout in PHP or otherwise.  The only thing I can think of that  
MIGHT be causing this if the memcached session keys are being  
overwritten due to high volume.  I took a look and we're currently  
allocating 512 MB to memcached.  We're getting almost 90,000 logins  
per day.  Though I can't tell exactly what our concurrent load is  
(though it'd be REALLY nice to be able to pull those kinds of stats  
out of Imp) I'd be willing to bet it's hovering around 3000-4000  
concurrent users at peak.

Could the situation I described with the kickoffs be attributed to  
load and memcached size?  If I added a second machine into the mix  
running memcached would that seamlessly distribute the load?  What  
would be a good memory size to allocate to memcached?  Also I'm  
noticing THOUSANDS of zero-length session files in /tmp of each horde  
front end.  Does Horde not clean up after itself?