Bridge interface performance

Hello.

I’m using VyOS 1.1.0 (Helium), installed on hardware, in a lab environment. It’s main function is that of a layer3 switch. I have two Cisco SG200-26 switches. Each cisco switch has two ports in an LACP-enabled LAG. The VyOS box has two dual-port NIC’s. In VyOS I have bond0 (GE20/21 on cisco switch 1) and bond1 (GE20/21 on cisco switch 2). I have several bridge interfaces and corresponding vif’s on both bonded interfaces. All connectivity is working, but I seem to be having performance issues with file copies between the two cisco switches with traffic on the same vlan. I’m using iPerf between two hosts for testing. I ran five, one minute tests, in each configuration and took their average to get the following numbers:

Server Client Speed

Switch1 Switch1 739 Mbits/sec
Switch2 Switch2 713 Mbits/sec

Switch1 Switch2 561 Mbits/sec
Switch2 Switch1 518 Mbits/sec

Did you check CPU usage on the vyos box during the transfer ?

Yep. I ran top during one of the tests and it was almost idle. As a sanity check I ran it again and it looked ok. Here’s the summary lines from about half way through another 5x 1 minute test.

top - 12:17:50 up 16:53, 1 user, load average: 0.09, 0.10, 0.07
Tasks: 82 total, 1 running, 80 sleeping, 0 stopped, 1 zombie
Cpu(s): 0.0%us, 0.0%sy, 0.0%ni, 88.2%id, 0.0%wa, 0.0%hi, 11.8%si, 0.0%st
Mem: 2028200k total, 184664k used, 1843536k free, 24824k buffers
Swap: 0k total, 0k used, 0k free, 85508k cached

This question may belong on a separate thread, but I’m trying to apt-get install ntop using the debian squeeze and squeeze-backports repositories and am having trouble with python dependencies. Does anyone else have ntop working on helium? and how did you go about it?

So…

Windows file copy reports expected speeds when traversing the two cisco’s across the VyOS bridge. I upped the TCP window size in iPerf and saw expected numbers. Still unsure why the default window size of 64KB bombed when using bridge, but not on same switch, but I don’t care about that now.