Dell Poweredge 2950 mystery, 1 year

by Duf Anything and Everything

I have been fighting with a Dell Poweredge 2950 server we have here for months.  We bought a couple of these servers and loaded them up with multiple processors, huge hard drives and tons of RAM.    However after the server was brought online we noticed that we were getting poor network performance.  Specifically, a process that pulled data across the network that used to take 2 or 3 seconds with an older, less powerful server now was taking nearly double the amount of time with the new lighting fast hardware.

So I did a Google search for “poor network performance Dell Poweredge 2950″ and was bombarded with results.  Tons of other people were having similar problems with bad network performance on their 2950’s as well.  Most of the information pointed the finger at the internal dual Broadcom NIC’s and specifically their TOE capability.  TOE is a feature in NIC’s that is supposed to increase network performance, well apparently in it’s implementation in the 2950 often causes more problems than good.

I read about several solutions to the issue such as  disabling the ” TCP Chimney”  by issuing the command “Netsh int ip set chimney DISABLED ”  as well as changing some registry entries to turn off the functionality.  I went as far as removing the TOE hardware key from the server to intentionally break TOE.  It made no difference, I still was seeing slower network performance on the new server.

So I decided to get drastic.  I ordered a dual port Intel NIC for the server and disabled the onboard Broadcom NICS.  I slapped the Intel NIC into the server, confident that this would finally eliminate my issue.  Imagine my frustration when I fired up the server, did a test and again saw similar results. WTF.

I already had tried the simple things like different cables, connecting directly to the switch and verifying connection settings on the switch.  The switch ports didn’t show any symptoms of a bad connection like dropped frames or corrupted packets.  

Well I was running out of ideas so I decided to get a new gigabit ethernet switch and connect the problem server to it as well as the server I am trying to copy data to and from.  I hooked the switch up, did my file copy test and STILL I see results that are worse than the 5 year old servers it is replacing.  

I’m very annoyed.  We have used Dell servers exclusively since I started down here and for the most part have had no problems with them.  However I don’t think I should have to jump through hoops to get a brand new server to at least perform on par with older hardware from a network perspective.

I guess it’s time for some more Google searching…

I just noticed in the “year ago post” that we had Buttons put to sleep one year ago today.  It instantly made me very sad.  That day was one of the worst I can remember.