Sorry, this is going to be a bit long - We have recently started having problems with IMAP connections dropping and clients having difficulty reconnecting. It affects a subset of IMAP users at any one time i.e. I might be experiencing it but the guy in the office next door may not. The severity of the effect is dependent on the client e.g. outlook, Eudora are badly affected; thunderbird less so and evolution not at all.
I have tried rolling back our 4 Windows Server 2003 NLB CAS servers to their pre-problem state to no effect. Checked all the usual things max connections etc. but that all looks OK. There are numerous errors and warnings in the event log but the pattern is the same as before the problems started and in any case they don't seem to tie in with my session when experiencing problems. During testing we found that connecting clients directly to particular CAS servers resulted in no dropped connections. In the hope that problems were being caused by a particular CAS server we tried them all but they were steady as a rock. Now my attention turns to NLB. Whilst experiencing this problem with outlook it goes offline and refuses to come online but if I telnet into port 143 on the NLB cluster address from the same client machine I get straight in. What's going on there? I have been researching NLB to see if there is anything about the way it works that could explain this. The 4 nodes of our NLB cluster handle OWA and IMAP connections (OWA is unaffected by the way). The port 143 and 993 rules are set to equal load and single affinity. It appears that whatever way NLB retains the session states a particular PC will always connect to the same node if it's available. If I stop the node my PC usually connects to, it will switch to another but as soon as the original node is active again it will switch back. I tried shutting my outlook session down for a period and stopping one of the other nodes. I then restarted the node and opened up outlook. I would have expected that outlook would grab a session via the newly restarted, more lightly loaded node but it just went via the same node it always uses. So it looks as though NLB 'remembers' the first node used by a particular IP address and will always use the same node thereafter if it's available. I'm thinking that we have been running this NLB cluster for 3 years with at most (and rarely) 2 nodes out of action at any one time. So - Does this mean we have built up a large volume of session state data? Could session state data be corrupted in such a way as to cause our IMAP problem? Does NLB store the session state in memory? Is there any way of clearing the session state data? Thanks if you've read this far! Clive McDowell Information Services The Queen's University of Belfast --- To manage subscriptions click here: http://lyris.sunbelt-software.com/read/my_forums/ or send an email to listmana...@lyris.sunbeltsoftware.com with the body: unsubscribe exchangelist
