Question

We are developing a client-server system where the client connects to a service and fetches an image from a buffer. The request runs at 25 hertz (25 requests per second) over a NetTcpBinding. The image data which is sent contains the image buffer (byte[]) and some meta data about the image.

What we are experiencing is that occasionally, the server does not respond for 5 seconds (5020 to 5050 ms), and we can’t figure out why.

Running svc logging on the client we see the following

Activity Boundary                   Suspend       10:00:00:000
Activity Boundary                   Resume        10:00:00:003
Received a message over a channel   Information   10:00:05:017

This occurs both when running the server as a managed WCF service, and an unmanaged WWS service

It can happen once every 100.000 requests, once per night, or several times per minute at seemingly random intervals.

Does anyone know what might cause this issue?

Answer 1

We found the solution buried in the Microsoft customer support database.

The 5 second delay is due to the firing of the SWS(Silly Window
Syndrome) avoidance timer. The SWS timer is scheduled to send the
remaining data which is less than 1 MSS (Maximum Segment Size, 1460
bytes) and the receiver is supposed to send an ACK advertising the
increased receive window and indicating that the remaining data bytes
can be sent. However, if the receiver sends an ACK when it can be
ready for sufficient buffer within 5 seconds, the SWS timer cannot
recover the 5 seconds delay status due to a race condition.

http://support.microsoft.com/kb/2020447

This issue only occurs when using localhost or 127.0.0.1. The delays do not occur when running the service and client on different machines.