I am using PG 9.0 on a Windows 2008 R2 server, and I have been seeing a lot
of TCP connection errors in the log files. With Wireshark I was able to see
two things. The first was checksum errors and the second was bad TCP
connections, which I believe the latter is a result of the checksum errors.
I have a Broadcom NetXtreme 57xx Gigabit Controller, which does not permit
me to change the checksum offload configuration setting. Therefore, instead
of using md5 in the pg_hba.conf file for authentication, I am now using
password (not encrypted). When I made this change, I had no connection
errors and the speed of the jobs increased by at least 4 fold.
Would changing the NIC fix this problem or could there be a bug in the PG
Windows software? Can anyone shed light on this topic for me and how I
might best address the problem?
I am not sure why I changing the md5 to password improved performance, but
it might be simply because PG is not trying to resolve the problem. Is
there some other piece I am missing that I need to consider.
Thank you for your suggestions,
Mike