iperf 2.0.14

[Robert McMahon]

Hi All,

I don't know if this is relevant to you or not and, if not, sorry for the spam. We're in an early field test EFT phase for iperf 2.0.14.  One of the new aspects is support for end to end read to write latencies, i.e. speed and capacity, which is a direct measurement of both.  More here

Thanks and my apologies if this seems like spam to you,
Bob McMahon

[Chris Ritzo]

Hi Bob,

Thanks for writing to share about the testing for iperf 2.0.14. We have considered the possibility of hosting iperf 2 or 3. I wonder if this is an appropriate thread to discuss the nuances of each and whether one or both would be useful to M-Lab users.

Thanks again,

Chris

[Robert McMahon]

Fine by me.  I do talk with the iperf 3 team and have some idea of their goals.   The number of 2 vs 3 might be confusing as people think a larger number suggests the same but better.  In this context, 2 vs 3 are completely different code bases. 

One technical thing that's been missing by my judgment is direct measurements of both capacity (peak average throughput) and speed (latency.)  Some use throughput/latency or network power as the metric to optimize.  Ping and packet latencies aren't necessarily good proxies for actual use experience.  I think the write-start to read-complete latency is much better.

Also things like connect times (TCP 3WHS) are a big deal too and should be measured.

We've added a lot to 2.0.14 per our semi-conductor customers.  We supply NICs, WiFi chips, switch fabric chips, etc. for all types of device and network equipment and this broad customer base has provided some inputs to the features implemented.

Bob