perfSONAR User Q&A and Other Discussion

[Hyojoon Kim <>]

Hi all, 

So I just found out that the perfSONAR 3.5.1 uses iperf3 instead of iperf by default when doing bandwidth tests (bwctl). This actually had a negative effect on the bandwidth tests that run on our low-cost hardware nodes, like our LIVA X boxes: the bandwidth performance degraded from ~900Mbps to ~780 Mbps.

I was trying to figure out why, and found this statement in the fasterdata website (https://fasterdata.es.net/performance-testing/network-troubleshooting-tools/iperf-and-iperf3/)

==

iperf3 thread model
In order to keep the code as simple and maintainable as possible, iperf3 is single threaded. This means that that you may be CPU-bound on some hosts, or on 40G/100G NICs. To run parallel stream iperf3 on mutiple cores, use the method shown in the table above.
==

Indeed, after the upgrade to v3.5.1, a single CPU in the LIVA box hits 99.9% CPU utilization when I run bwctl tests manually. When I force the bwctl test to use iperf instead of iperf3 (adding "-T iperf” option to bwctl command), the performance goes back to normal (~900Mbps), and both CPUs are utilized. 

Is there a way to make the regular bandwidth tests to use iperf instead of iperf3 by default in perfSONAR v3.5.1? 

Thanks, 

Joon