[horde] still having problems with dropped sessions

[Majeed]

Im still in the process of planning our new webmail system based around 
memcached sessions. Maybe others can correct me, but I have a hard time 
seeing 1gb of memory for session data for 3600-4000 concurrent mail 
users even being close to enough. Even if each user, during peak times, 
had .5mb worth of session data, it would still be quite a bit more than 
1gb. However, I'm not an expert. Yet :)

Good luck

Kevin Konowalec wrote:

>Horde: 3.1.2
>Imp: H3 (4.1.2)
>Turba: H3 (2.1.1)
>PHP Version: 4.4.1
>
>So I've added a second memcached instance of 512MB (for a total of  
>1024MB) but we're still seeing sessions being dropped fairly  
>frequently (and no messages in the logs are even remotely helpful).   
>By "dropped" I mean the user will be logged in and reading or writing  
>mail and suddenly he/she will find themselves booted back to the  
>login screen without explanation.  We have roughly 3600-4000  
>concurrent users at peak volume - is 1Gig enough memcached size?  And  
>if it's not a memcached problem, where else can anyone suggest we  
>look as the cause?  Since this is the busiest time of the year I  
>suspect that it's a function of load... but I'm not sure which part  
>of the system is failing because of it.  And it's strange in that  
>sometimes you can leave yourself logged in all night and your session  
>will still be there (as you'd expect) but other times you can be  
>booted out several times repeatedly without warning.
>
>I'm not sure if it's an indicator of the way students use the system  
>(ie logging in to check their mail several times a day) or if it's a  
>result of this repeated login problem but my log analysis showed that  
>the number of distinct users logging in per day is dwarfed by the  
>total number of logins per day at a ratio of about 3:1 and growing.
>
>
>
>  
>