2010-11-18 09:47 AM
I was troubleshooting some network performance issues and during my testing, came across some very interesting behavior that I cannot explain.
I wanted to see if anyone else had seen anything like this?
1. I setup one of the network interfaces directly from the filer to our core switch (1g, full, jumbo frames)
2. I tested copying large files 8gb using CIFS and was getting some 40mb/s performance (2 minutes to copy the file)
3. I then deleted this configuration and created a virtual network interface and added only one nic to the trunk. (same nic actually)
4. I tested the throughput copying the same 8gb file and got 20% faster network performance!
I repeated this test three times with the same file and got the exact same results.
Setting up as a VIF with one nic in the trunk was faster than direct nic connection.
This seems to fly directly in the face of logic.
Anyone got any ideas on this behavior?
2010-12-07 06:03 PM
Did you also change the configuration of the switch at all to match the configuration of the VIF? Did you set it up as single, multi, or LACP?
I'll try to recreate this in my lab.
2010-12-07 06:12 PM
I tried is as both single and multi on the switch.
That seemed to make no difference.
Which probably makes sense sense because I only added one filer interface to the VIF.
I discovered this purely by accident.
I had setup a VIF with three nics and was testing the performance as I unplugged the filer interfaces that were part of the VIF.
Then I compared performance copying large files using CIFS.
I compared it to a copy of a directly attached nic.
I deleted and recreated the VIF multiple times and tested performance against the single nic and it was always better.
2010-12-08 08:19 AM
Unfortunately, the FAS3050 is now in production and the network interfaces are no longer available to use for reconfiguration and testing.
I might be able to try it on a FAS3020 that is not currently in use at a future date.