perfSONAR User Q&A and Other Discussion

[Mark Feit <>]

(Sorry if the formatting is horrible; thank Outlook for it.)

Tim Chown writes:
    
➢ When we explain what perfSONAR is, we do get the question “so can we just 
run one-off iperf tests directly against it?”, which is not an unreasonable 
question. 

I can interpret that question three ways:

The first is “can I measure throughput from a non-perfSONAR system to a 
perfSONAR system just by running iperf?”  The answer to that is no.  Despite 
the changes we made in 4.0, that’s the same as it ever was; BWCTL didn’t 
start the iperf server until it was time to do a measurement, either.

The second is “can I make ad-hoc throughput measurements between perfSONAR 
systems without either being part of a mesh with repeating testing and 
everything that comes with it?”  The answer to that is yes and, again, is the 
same as it always has been.

The third is “can I make an ad-hoc throughput measurement from a perfSONAR 
system to a non-perfSONAR system that has iperf running in server mode?”  The 
answer is a qualified yes.  You can just run iperf directly, but you run the 
risk of distorting the results of measurements being made by pScheduler or 
having those measurements distort yours.  Of course, if you’re just running 
iperf, you can install and run that from pretty much any system you want and 
not involve perfSONAR unless you’re worried about having off-the-books 
throughput testing causing link congestion and/or influencing your results.

pScheduler has two things that can help with the third one.  One is a “pause” 
command that will cause any scheduled measurements not to be done until the 
“resume” command is issued.  (Already-running measurements will still 
continue to completion, and there is a way to tell it to automagically resume 
after a specified period.)  The better solution is a test called “idleex,” 
which does absolutely nothing and prevent other schedule-exclusive tests like 
throughput from running at the same time.  Try this at the command line:

        pscheduler task idleex --duration PT5M
 
Once the run of the task has started, you’ll have five minutes to run iperf 
separately and be assured of no interference.


➢ But having as clean a way as possible to allow direct iperf tests might be 
useful, and might help people understand the value of deploying a full 
perfSONAR system at their site.

The cleanest way possible is to not have perfSONAR in the picture at all:  
install iperf and start it in server mode.  If you really want to drive home 
why having perfSONAR is valuable, arrange a test manually, because 
eliminating that step and making measurement available on-demand all the time 
is why perfSONAR came to exist in the first place.

--Mark