perfSONAR User Q&A and Other Discussion

[Brian Tierney <>]

In general people trust TCP to "do the right thing", and back off, causing minimal impact on other traffic.

Since perfSONAR only runs 1 test a  time, 1 additional TCP flow should have minimal impact on other traffic.

Another option is to use the Linux 'tc' tool to throttle traffic from your perfSONAR host.

http://fasterdata.es.net/host-tuning/packet-pacing/

On Wednesday, March 11, 2015, Jeff White <> wrote:

So when doing a bandwidth test PS will obviously try its hardest to fill the pipe as much as it can.  Assuming the PS host can fill a link to that connection's full capacity and no limitations (traffic shaper, ACL, firewall, etc.) are in place, how does that affect other traffic using that link?

Take this example:

* PS node A in building A connected to core router A at 10Gb/s

* Building A's core router connected to building B's core router at 10Gb/s

* PS node B in building B connected to core router B at 10Gb/s

* 2Gb/s of normal traffic is flowing between core router A and B

* The PS nodes begin a throughput test between A and B

What happens?

The routers are not looking at QOS or whatever so they wouldn't know that the PS test is less important traffic and therefore should be dropped when capacity is reached.  So wouldn't the router just start dropping packets somewhat randomly, including packets of non-test traffic?

Are any users of PS putting limitations in place to prevent this?  If so, what and how?

--
Jeff White - GNU+Linux Systems Administrator
University of Pittsburgh - CSSD

-- 
Brian Tierney, http://www.es.net/tierney
Energy Sciences Network (ESnet), Berkeley National Lab
http://fasterdata.es.net