|
> Thanks for the response. OK. I can see a lot of
> connections in the
> FIN_WAIT state. The cluster is now lightly-loaded - but
> despite that the
> number of connections in the FIN_WAIT state seems too high? Is this
> something to worry about or is it normal?
FIN_WAIT timeouts are very high in linux. It's normal.
> When the cluster comes under a lot of load the
> InActConn numbers grow very quickly and jump to over 2000. Is
> there a resource penalty
> for FIN_WAIT connections? - the simple answer may be there is
> not enough
> capacity in the cluster to handle the higher loads.
Someone (Julian, Ratz?) a long time ago posted the cost per connection for
LVS. I believe it was 128 bytes per connection. So if you have 20
realservers each with 2000 connections LVS will take
Number of connections * 128 bytes = memory used by LVS.
Number of connections in this case is 20*2000, approximately. So 40,000.
40,000 * 128 bytes = 5,120,000 bytes, or ~5.1MB used by LVS for 40,000
connections.
Note that these figures don't include iptables calculations.
Regards,
P
|
