Arun, This is just a WAG, but I think that the problem still lies in the client side and how the load balancer is handling the requests. I would still be interested in seeing the netstat -atunep output from one of he affected machines when there is a high level of connections in a closed state. Even though the sessions aren't sticky, is it possible that the load balancer is handling keep alive in such a way that it's sending the back to the original server but on in a new state? (that's a little beyond my knowledge though).
Gary Smith From: Arunkumar Janarthanan [mailto:arunkumar.webad...@gmail.com] Sent: Wednesday, January 27, 2010 11:37 AM To: users@httpd.apache.org Subject: Re: [us...@httpd] Apache Server status page shows uneven number of requests across load balanced web servers Thanks Gary, The load balancing method we use is weight based, so apparently the load balancer suppose to send the traffic to the web server has least number of connections. However the Netscaler support confirmed they see eventual number of traffic routed to all 4 web servers. Regarding the Apache restart time, I restarted all 4 servers same time and checked could still see the requests on 1 & 3 are being handled. I dont think we have sticky session enabled at loadbalancer level. For the same problem we have not before changed the load balancer settings from sourceIP persistence to Weight based to have the load distributed equal on all the servers. IP port Type State Req/s 731-...30_80 x.x.x.x 80 HTTP UP 16/s 731-...31_80 x.x.x.x 80 HTTP UP 16/s 731-...31_80 x.x.x.x 80 HTTP UP 14/s 731-...55_80 x.x.x.x 80 HTTP UP 20/s Rsp/s Reqb/s Rspb/s ClntConn SvrConn MaxConn SurgeQ 731-...30_80 16/s 12406/s 345117/s 7 285 0 0 731-...31_80 16/s 10956/s 365025/s 7 262 0 0 731-...31_80 14/s 11140/s 368655/s 6 268 0 0 731-...55_80 20/s 14322/s 420105/s 7 272 0 0 However the Apache server status still shows the number of requests being handled huge difference. I have attached the balancer-manager page as well with this email. Best Regards, Arun J
