Intermittent interrupt issues of ISA Server network load balancing

Intermittent interrupt issues of ISA Server network load balancing

  • Comments 2
  • Likes

Last week, I encountered a strange issue for network intermittent interrupt.

My customer deployed two ISA Servers as a NLB array, but the network traffic to NLB array is intermittently interrupted periodically after a few minutes.

For example, if I pinged the NLB virtual address or NLB node dedicated address continuously, the output was:

Pinging 10.10.9.2 with 32 bytes of data:

Reply from 10.10.9.12: bytes=32 time<1ms TTL=127

...

Reply from 10.10.9.12: bytes=32 time<1ms TTL=127

Reply from 10.10.9.12: bytes=32 time=1ms TTL=127

Reply from 10.10.9.12: bytes=32 time<149ms TTL=127

Reply from 10.10.9.12: bytes=32 time<241ms TTL=127

Reply from 10.10.9.12: bytes=32 time<213ms TTL=127

Reply from 10.10.9.12: bytes=32 time<234ms TTL=127

Request timed out.

Reply from 10.10.9.12: bytes=32 time<1ms TTL=127

Reply from 10.10.9.12: bytes=32 time=1ms TTL=127

...

Reply from 10.10.9.12: bytes=32 time<1ms TTL=127

Reply from 10.10.9.12: bytes=32 time<156ms TTL=127

Reply from 10.10.9.12: bytes=32 time<202ms TTL=127

Reply from 10.10.9.12: bytes=32 time<212ms TTL=127

Reply from 10.10.9.12: bytes=32 time<190ms TTL=127

Request timed out.

Reply from 10.10.9.12: bytes=32 time=1ms TTL=127

Reply from 10.10.9.12: bytes=32 time<1ms TTL=127

...

After analyzing captured packets, it seems sometimes there is a big delay and timeout from client to NLB array.

If I disable NLB, everything is okay.

This is strange, and my troubleshooting steps included:

  • Disabled EnableTCPChimney, EnableRSS, EnableTCPA and DisableTaskOffload in Windows Server 2003, but this didn’t help;
  • The NIC mode is “Broadcom BCM5708C NetXtreme II Gige”, and I updated its driver to newest one, but it didn’t help;
  • Disabled all offload functions in NIC driver’s advanced features, but it didn’t help;
  • Reviewed ISA Server’s configuration - everything is properly configured.
  • Asked network devices vendor to review switch’s configuration, but the switch’s configuration seems okay.

This was starting to make me crazy and I thought I didn’t miss anything.

 Finally, I tried to disable all of the advanced features in NIC driver configuration, beside all offload functions, I also disabled Flow Control, Ethernet@WireSpeed, Interrupt Moderation.

After that, everything was okay!!!

clip_image001

This is just a hint for you. Happy to help.

Meibo Zhang, Premier Field Engineer

Comments
  • The problem was most likely the Ethernet@wirespeed setting

    Have a look here: www.pcreview.co.uk/.../thread-1510997.php

    The likely cause is that the way the quality of the cable is assessed disrupts the active connections

  • Thanks a lot for your notice.

    I have no chance to test which option is the root cause in that time, and I will pay attention to it in future.

Your comment has been posted.   Close
Thank you, your comment requires moderation so it may take a while to appear.   Close
Leave a Comment