We have an ASP.NET MVC 1.0 application running on Windows Server 2008 – Standard (32 –bit), Dual Core Xeon (3.0 GHz), 2 G.B R.A.M. Most of the times application renders response in 3-4 seconds, but sometimes users get very late response and delay is up to 40 seconds or more than a minute.
It happens in following way: User browsed a page, idle for 5, 10 or 15 minutes, tried to browse same page or some other. Now, there is a chance that he will see late response whereas the app pool is still up and running. This can happen with any arbitrary page.
We have tried followings/observations.
Moved the application to stand alone web server
App Pool idle shutdown time is 60 minutes. There are no abrupt shut downs/restarts.
CPU or memory doesn’t spike.
No delays in SQL queries.
Modified App Pool setting to run in classic-mode. It didn’t help.
Plugged-in custom module to log all those requests which took more than 5 seconds to complete. It didn’t pick any request of interest.
Enabled ‘Failed Request Tracing’ to log all those requests which take 20 or more seconds to complete. It didn’t log anything.
Event Viewer, HTTPER log, W3SVC logs or WAS logs don’t indicate anything. HTTPERR only has ‘_ _ Timer_ConnectionIdle _ _’ entries.
There is not much traffic to server. This can happen also if only two users are active.
Next we captured TCP/IP terrific on both a user and server end with Wireshark and below are details in brief of this slowness:
Browser sends a request for ~/User/Home/ (GET Request) by setting up a receiving end point using port 'wlbs(port-2504)'. I'm not sure if this could be a problem in some way that browser didn't hand-shake with the server first and assumed that last connection is still open, whereas, I browsed the same page 4 minutes ago and didn't perform any activity with site after that. If I see the HTTPERR log, it indicates that it has ‘_ _ Timer_ConnectionIdle _ _ _’ entry for my last activity with server.
Browser (I was using Chrome) waits for any response from the server, doesn’t find any then starts retransmitting the same request using same end point after incrementing wait intervals, e.g. after 8, 18, 29, 40, 62, and 92 seconds.
All these GET requests were received by server as well. But, server didn’t send any packet to client.
Browser didn't see any response on the end point it set up in point 1, it opened a new end point 'optiwave-lm (port-2524)', did a hand shake with the server and transmitted the same request again. Server received, processed it, and returned successful response.
What happened to earlier 6-7 requests? Whether they were passed on to HTTP.SYS or not? Why Failed Request Tracing not logged anything, we didn't find any clue yet. Server served the same page successfully just 4 minutes ago.
Looking forward for more suggestions/solutions.
-- Thanks
