|
On Wed, 2007-07-18 at 09:49 +0200, Sebastian Vieira wrote:
> So i finally got the two directors off of vmware, onto identical physical
> (compaq ml370) machines with shiny new dualnic cards, oh i'm so happy.
Thanks $deity for that :)
> Except that now all of a sudden, RDP connections that have been idle for
> more than 5 minutes (the default ipvsadm --set timeout?) get disconnected.
> And this is what i expected 2 years ago when i first put it into production.
> But these timeouts never occurred. Clients could stay connected all day,
> doing nothing, and LVS wouldn't drop the connection. Now it suddenly does
> and even not with every connection, which is extra strange (to me). The
> things that changed are:
>
> - newer kernel
> - physical hardware instead of vmware
> - other NICs (dual intel 1 gig cards)
> - channel bonding
>
> The LVS configuration didn't change at all, just the heartbeat which is
> (temporarily) sent over the same line as the RDP traffic.
>
> Is there a logical explanation for this, or is LVS now working as intended
> and i was just lucky all this time? :)
Good question!
When you say "the clients are idle", do you really mean "idle"? In LVS
terms that would be determined by whether or not they exchange any
traffic. If they don't, then I can understand the connection timing out
perfectly well; if they do, further work is needed.
Can you run tcpdump (selectively!) on one established session which you
then leave "idle" (your meaning of the word) and see what the session
does over time?
Thanks
Graeme
|
