Hi, Malcolm
Thanks for usefull infomation.
> It doesn't take up much in the way of resources, although I seem to
> recall some discussion that this had recently raised from 128 bytes to
> 256 per entry?
It's good information that 'FIN_WAIT' entry doesn't consume much memory.
> That makes more sense TIME_WAIT is not usually seen much.
> FIN_WAIT means that the TCP connection has timed out (2 mins) but LVS
> will keep it in the table
> for the default timeout of 15mins.
I think FIN_WAIT means LVS waiting for FIN to sent from real server, because
LVS can't see FIN packet from real server in DSR topology.
In viewpoint of TCP, FIN packet should sent from both client and server.
The FIN_WAIT is not clearly finished state nor timed out state, but in DSR
topology, LVS should delete this entry because LVS can't see packet from real
server to client.
And 'FIN_WAIT' entries remains for 120sec by default
[root@lvs1 ~]# ipvsadm -L --timeout
Timeout (tcp tcpfin udp): 900 120 300
[root@lvs1 ~]#
And after one request finished, FIN_WAIT remain 'tcpfin' duration in table.
I checked it by 'watch -n 1 "ipvsadm -L -c -n"'
* expire column indicate countdown timer for deletion of entry.
Is my understandig correct?
Taku.
>
> It doesn't take up much in the way of resources, although I seem to
> recall some discussion that this had recently raised from 128 bytes to
> 256 per entry?
>
>
> 2008/8/14 taqu <taqumd@xxxxxxxxx>
>>
>> Hi,
>>
>> Sorry , I noticed big mistake in last mail I sent 'TIME_WAIT
>> connection on DSR mode.'
>> All ''TIME_WAIT' should be replaced to 'FIN_WAIT' !!
>
> --
> Regards,
>
> Malcolm Turnbull.
>
> Loadbalancer.org Ltd.
> Phone: +44 (0)870 443 8779
> http://www.loadbalancer.org/
>
> _______________________________________________
> LinuxVirtualServer.org mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
> Send requests to lvs-users-request@xxxxxxxxxxxxxxxxxxxxxx
> or go to http://lists.graemef.net/mailman/listinfo/lvs-users
>
|