I am running Server 2008 R2 Standard on a Dell 2950, the server is being used as a terminal server but is experiencing very poor network performance, to the point where I have now had to revert back to Server 2003 Standard (which doesn't have these problems) until I can work this out, the server does need to be upgraded to 2008 due to software we run not working properly on 2003. We only have the one 2008 R2 server, everything else is 2003, XP Clients, Wyse Clients, all are on gigabit network, there are only about 8-10 users on this server but applications are often not responding. There are heaps of posts about how people magically fix this sort of problem, but none of them seem to work for me. As an example here are some indicative figures I got when testing the server, I was the only one on the network and this was a fresh install of 2008 R2, latest drivers/firmware from Dell/Intel, and all the latest Microsoft patches, the NIC is a Intel Dual PT gigabit setup in a team, I have also tried the onboard Broadcom gigabit with no luck. Copying files from 2008R2 -> 2003, the speed varies from <1MB/sec ->~14MB/sec. Copying files from 2003 -> 2008R2, the speed varies from ~25MB/sec -> ~100MB+/sec Copy files locally on the 2008R2 server is very fast What I have tried: - various netsh settings to disable chimney/rss/autotuning(to various options) (there are plenty of posts on the web about this) - disabled SMB2.0 via registry - tried last 3 versions of Intel drivers + heaps of Broadcom ones - tried fresh installs of 2008 SP2, and 2008R2 - both have same problem - purchased Intel NIC - tried changing every setting I could find on the Intel NIC - tried direct connection to my laptop (XP) - Swapped cables, switches The other day after setting up the teamed Intel (for about the 20th time) it magically started copying really fast to the first server I tried, but all other servers I tried copying to were still slow (weird), the 800MB file I had been using to test took about 7secs to copy in each direction, where as before it was around 120secs out/45secs in (and on 2003<->2003 its about 20secs each direction). I rebooted the server and it was awful again to all the servers I tried. Does anyone have any thoughts on what I can try? Any help is much appreciated. Thank you

Have you tried disabling "TCP Large Send Offload" by any chance? As per http://social.technet.microsoft.com/Forums/en-US/winserverhyperv/thread/bdc40358-45c8-4c4b-883b-a695f382e01a

This was one of the many settings I tried on the Intel NIC, it didn't help. Thanks for the input though.

I found a forum where someone said adding QoS Packet Scheduler to the 2003 servers fixed their problem, I tried this and it seemed somewhat successful, but there are still issues, does this give anyone an idea of what might be needed? I couldn't test it on my 2008R2 box as this is currently running 2003 again, however I had a 2008 SP2 box that was having the same problems, I added QoS Packet Scheduler to 4x 2003 servers and tested the copying speed, it went from ~2MB-8MB/sec outbound (2008->2003) to consistently reaching >100MB/sec, however after further testing one of those four servers has dropped back down to ~2MB-8MB/sec. I did notice that after initially making the change the copy speed started off slow and pickup to >100MB/sec after about 5-10 secs the first time, thereafter it was >100MB/sec on subsequent copys. Any thoughts?

I finally did some further testing on this, the QoS didn't solve anything as all the server connections became slow again after a reboot of the 2008 server. I've logged a call with Microsoft to have a talk to someone about this, I'll post back the result.

MS Technician is looking into this problem for me, so far he has asked me to do the following: 1. Disable RSS, Chimney, AutoTuning on 2008R2 2. Disable the same on 2003 via the registry as listed below. a. Click "Start"->"Run", type "regedit" (without quote), click "OK". b. Locate the following registry subkey: HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\Tcpip\Parameters c. Right-click "EnableTCPChimney", and then click "Modify". d. In the "Value data" box, type "0" (without qutoe), click "OK". e. Right-click "EnableRSS", click "Modify". f. In the "Value data" box, type "0" (without qutoe), click "OK". g. Right-click "EnableTCPA", click "Modify". h. In the "Value data" box, type "0" (without qutoe), click "OK". i. Exit Registry Editor, and then restart the computer. 3. Install http://hotfixv4.microsoft.com/Windows%207/WindowsServer%202008%20R2/sp1/Fix312999/7600/free/413994_intl_x64_zip.exe None of these made any difference for me, I am in the process of sending him Network Monitor log files from 2008R2 and 2003SP2, also sent them MPS Report as requested. I will post back any further developments.

MS had me install various updates to do with tcpip.sys and srv.sys on 2008R2/2003, further testing seemed to result in the slow network speed going both ways to/from Server 2008R2, Server 2003 remained fine. They asked me to do further testing on the switches, the result at this stage seems odd to me, but this is what I have found after extensive testing (maybe it will help someone else one day). My Cisco 800 Series router (which is the default gateway) is causing Server 2008R2 to perform slow on the network but not the 2003 servers, i.e. Test setup: HP or Dell switch, 2x 2003 SP2, 1x 2008R2, 1x Cisco 800. Fresh restart of everything it seems to run ok for a short time then it slows down, when it slows down the following happens, when copying the test file from 2003 - 2003 from 2003 console it is fast, when copying test file from 2003 - 2008 or 2003 - 2003 from the 2008R2 console it is slow. During the slow copy if I unplug the Cisco 800 it seems to pause for 10-20 secs, then starts copy really fast, and continues to until a short time after plugging in the Cisco 800 again. If it copies slow and I cancel it, pull the Cisco 800 out and then start the copy again the following occurs: small delay and the copy screen shows "Discovering items..." then it copys fast until I plug the Cisco 800 back in. At this stage I'm going to wait to see what MS have to say about the 2008R2 and my router, I will look into getting the latest firmware for the Cisco, and test an old Snapgear/Request device I used to use prior to the Cisco. I'll post back what when I have further information.

Two issues that I have encountered through the years that attribute to slow network performance: 1) Driver Issue (check the vendor's site for updated drivers) 2) Computer NIC/Switch Port Speed/Duplex Mismatch. (verify the speed/duplex setting on computer and switch port) Visit: anITKB.com, an IT Knowledge Base.

Thanks for your suggestions, I have already got all the latest drivers for the NICs and checked the switch/nic settings. I replaced the Cisco 800 with my very old SnapGear Lite2 and Request 531R gear and haven't had any further problems, it's only been a few hours but before now I've been able to replicate the issue easily within a few minutes, I'm just about 100% sure that the problem has been resolved and it was something to do with the Cisco 800. I will update the firmware on the Cisco 800 when I have time and test it again. This is what MS had to say about the issue: I just checked the network trace file again, and cannot find any clue that might be related to the router device. For example, in “10.0.29.22 2008R2 Server.cap” file, after excluding network communication between Win2008R2 and the Win2003 server, there is only some broadcast (browser), SNMP, and SMB (to SSERVER) traffic. I believe these traffic would not cause the poor performance issue. According to the test result, the possible cause might be the “Next Generation TCP/IP Stack”. It is by design that Windows 2008 (R2) systems use the “Next Generation TCP/IP Stack” which is a complete redesign of TCP/IP functionality. Technically, the “Next Generation TCP/IP Stack” is designed to improve network performance and efficiency. However, some network device cannot handle such high efficiency network traffic, and therefore it causes the low performance issues due to data loss and retransmit. Based on current status, I would suggest we involve the hardware vendor to ensure if it is a hardware issue. We can also enable port mirror on switch devices to see whether the data loss occurred on the hardware.

I don't really understand why this wasn't an issue before I installed the 2008R2 servers, however the problem was: The internal interface on the CISCO was configured with: ip local-proxy-arp once I disabled that it worked fine.