Get Those Checksums in Check

As computer components and software increasingly get better and faster, it just makes sense to virtualize some of the internal infrastructure of your business.  The ability to run ten separate servers on a single physical box is just too good to pass up, right?  It is if you put the right amount of planning and resources into the project.  We have seen many customers move to a virtual environment over the past few years and we expect more to do so in the future. In an effort to help increase the performance of the Escape Online5 application servers running in virtual environments, Escape is always listening to our customers and identifying known issues which can affect the system.   Recently, we have come across a very common issue among Windows servers running on VMware VSphere 4.0 and 4.1.  This issue involves the offloading of tasks to the network interface card.  While this may be a good thing on physical network adapters, it may not be so good on virtual servers that have virtual NICs bridged to real physical NICs.

Escape used Wireshark (a well known packet monitoring tool) to monitor the network interface on many of our virtualized application servers running on VSphere 4.1 and Windows 2008 and Windows 2008 R2.  We noticed that within a three minute period, there were over 4000 checksum errors on most of these systems.  Checksum errors occur when the packets sent over the network don’t match from one end to the other.  This was shocking, especially since the systems we monitored were not very busy with client requests.  After some investigation, it was discovered that VMware is aware of this issue and has a fix noted in the following KB article.

This article describes how to disable the Task Offloading on the TCP network interface.  Once this is done, most, if not all, checksum errors should be eliminated.