network speed issues

[CESIL]

Hi All

 

I have been running jperf to test the raw speed of our network and have found an anomaly...

 

I ran the test on two identical Dell 760 PCs (I checked the nic settings and they are also identical) and one showed a speed of about 800Mbps the other about 400Mbps.

 

They are in the same room connected to the same switch and I even tried the cable from the faster machine on the slower and it made no difference.

 

...so, the question is, what do I look at to determine the reason for the speed difference?

 

The speed difference has been the same every time I have tried the test...

[sted]

driver version. switch port. depends exactly how your testing hdd speed. firmware or bios revision. ones downloading wsus or other updates

us

[CESIL]

driver version. switch port. depends exactly how your testing hdd speed. firmware or bios revision. ones downloading wsus or other updates

us

Drivers are the same version.

HDD is not being tested...jperf is doing a tcp transfer point-to-point in memory (AFAIK)

I will look at firmware and bios and try the slower PC on the same socket as the faster one...

[Michael]

Are the hard disks themselves the same make/model/spec? 7200rpm vs 5400rpm for example does make a difference.

[FN-GM]

Same BIOS version?

[Michael]

Another possibility is the SATA speed. Depending on the age of the motherboard, some have dual SATA 150 and SATA 300 ports. Can you confirm the hard disks are connected to the same SATA port? Typically they're coloured differently to indicate the difference in speed.

 

I do suspect (as you can tell) this is specifically a workstation/hardware issue rather than a network issue.

[twin--turbo]

Can you tell if the switch is running Full Duplex on the port?

 

Rob

[CESIL]

OK, so I have now verified that the hardware, BIOS drivers etc are all identical and set the same.

 

The switch ports are running the same full duplex, 1000Mbps.

 

Yesterday I connected the computers together to bypass the network and got an interesting result.

 

The slower machine acting as server for the test receives at 900Mbps plus, but when acting as client runs at 4-500Mbps.

 

So...I guess that means a flaky nic? Or does it?

[snoerre]

For Iperf on Windows you have to use some options in order to get more performance:

 

Network Throughput and Performance Guide - Xen

 

# on receiver

iperf -s -f m # add "-w 256K -l 256K" when sender or receiver is a Windows VM

 

# on sender

iperf -c -f m -t 20 # add "-w 256K -l 256K" when sender or receiver is a Windows VM