Hmm. Now I'll have to do some tests at work next week using various combinations of OS and VMware ESX/vSphere. I'll get back to you. :)<br><br><div class="gmail_quote">On Fri, Apr 16, 2010 at 3:36 PM, Adam Thompson <span dir="ltr"><<a href="mailto:athompso@athompso.net">athompso@athompso.net</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">FYI. Windows networking sucks, across all versions of Windows. (At<br>
least, at Gigabit speeds, it does.)<br>
<br>
Between my Windows 7 PC<br>
(3.06Ghz HT Pentium 4, Intel PRO 1000/MT Desktop Adapter [82540EM])<br>
and Gentoo Linux server<br>
(2 x 2.2GHz Xeon [P4 core], PCI-X [64-bit,66Mhz] Intel PRO 1000/MT<br>
Dual-Port Server Adapter [82546EB])<br>
<br>
Using "iperf -s" and "iperf -c 192.168.x.x", I get - consistently -<br>
between 238Mbps and 250Mbps **in both directions**.<br>
<br>
Interesting to note is that the Windows 7 PC is CPU-limited on both<br>
sending and receiving, despite having all the offload and acceleration<br>
features turned on. Simultaneously, the Gentoo box sees a spike on one<br>
CPU... to 1%.<br>
<br>
In UDP traffic, however, I see a startling asymmetry: PC-to-Linux can only<br>
send UDP at ~50Mbps, Linux-to-PC sends UDP at ~475Mbps. WTF?<br>
<br>
I also tried the Parallel option, which runs multiple streams in parallel.<br>
Results are consistent, aggregate bandwidth is still only about 260Mbps.<br>
Two streams pegs both logical (HT) CPUs on the client at ~95% utilization.<br>
<br>
I tried enabling and disabling both the Windows Firewall and my AV<br>
software, which made essentially no difference at all.<br>
<br>
Then I also tried running the same set of tests from a Windows 2003 R2<br>
server<br>
(2 x 2.13GHz Athlon MP, PCI-X [64-bit,66Mhz] Broadcom NetXtreme<br>
[BCM57xx])<br>
N.B. this is the same model of motherboard Kevin is using in his Linux<br>
server.<br>
and observed slightly better throughput to the Linux server: about<br>
380Mbps. Still better was Linux-to-W2K3R2, at about 440Mbps. UDP<br>
performance was similar to Kevin's results, at 103Mbps W2K3-to-Linux &<br>
950Mbps Linux-to-W2K3.<br>
<br>
CPU utilization on the W2K3R2 box exhibited similar properties as the Win7<br>
client, although none of the tests actually managed to max out either CPU,<br>
instead peaking at about 70% utilization.<br>
<br>
Just for kicks, I ran a test between a RHEL5 VM (under VMWare ESXi 4.0U1,<br>
VMXNET adapter) and my Linux server... 920+ Mbps every time! Of course,<br>
that number may be suspect since the UDP test reports speeds of 45.3Gbps<br>
:-).<br>
<br>
Conclusion: the culprit is the Windows networking stack. Note that W2K3R2<br>
is supposed to support zero-copy TCP offload, theoretically allowing<br>
network communications at wirespeed with extremely low latency. So much<br>
for that idea...<br>
<br>
-Adam Thompson<br>
<<a href="mailto:athompso@athompso.net">athompso@athompso.net</a>><br>
(204) 291-7950<br>
<br>
<br>
<br>
_______________________________________________<br>
Roundtable mailing list<br>
<a href="mailto:Roundtable@muug.mb.ca">Roundtable@muug.mb.ca</a><br>
<a href="http://www.muug.mb.ca/mailman/listinfo/roundtable" target="_blank">http://www.muug.mb.ca/mailman/listinfo/roundtable</a><br>
</blockquote></div><br>