Hi, On Tue, Jan 11, 2011 at 01:25:43PM +0530, Paras Fadte wrote: > Hi, > > In the attached screenshot ,
Please avoid JPEG screenshots. This one is barely readable. You can save your stats page in HTML, it's all self-contained. And if you prefer to hide some info from there, then save in a lossless image format such as PNG or GIF. > wanted to know If the backend current queue of > "3069" shown under "log_servers" section is caused due to saturation of the > defined backend machines logserver4/5/6/7/8. Yes, only a situation where servers are saturated can cause the backend to queue requests. In your case, since we're seeing a current connection count around 500 with a limit at 20000, I suspect that you have enabled dynamic limits with a minconn and a maxconn on each server. In your case it seems to protect your servers well, because you have 50% of the requests immediately served and 50% in the queue. Assuming the requests are equally drained by all servers, you'd have 3069/5 = 614 requests to be processed for each server before a new incoming request gets served. At 790 req/s/server, that's around 780ms spent in the queue waiting for a server to become available. The total response time for the client is around 780+650 = 1.4s. You should try to play with those parameters to see if your servers respond faster with more requests or with less. Often you'll notice that your servers respond faster with less concurrency, and it will make sense to lower the limits so that they drain requests faster. > How is this queue different > from the backend queue shown under "HA-Proxy_LB" ? I suspect that HA-PROXY_LB is just a "listen" instance you have declared to accept the traffic and to send it to "log_servers". Only its frontend part is used then and it has no server, so it will never queue anything. The queueing only happens in the backend. Hoping this helps, Willy