perfSONAR User Q&A and Other Discussion

[Mark Feit <>]

pete writes:   
    
    Does anyone have any experience or insight into how this might effect 
    our perfsonar regular tests?

The general rule is that you should adopt whichever value or range is more 
generous.

The local port range on the perfSONAR machines is the default; adding to it 
just expands the range of ports usable as source ports for network 
connections.  If the speed tester is all HTTP(S), you shouldn’t need to 
change this value unless you’re anticipating a very large number of 
connections.

The congestion control algorithm is a personal-preference thing.  HTCP may 
work better for the speed tester’s payload and might net you better 
throughput with perfSONAR.  All I can advise here is to benchmark both things 
with both settings and see how they do.

You should also be aware that perfSONAR won’t be aware of what the speed 
tester is doing and vice versa, so testing throughput with both at the same 
time may result in distorted results.  As discussed recently in another 
thread, we are looking to modify an existing speed tester so it can integrate 
with pScheduler and schedule a timeslot for its activities.  Let us know how 
that one works out for you.

Hope that helps.

--Mark