perfSONAR User Q&A and Other Discussion

[Alex Moura <>]

Hi Matt, 

Thanks for the quick reply, which confirms what we thought about how it should work, regarding the TCP negotiation of the lower MSS for transmission.

On a side note, if there is a need to customize individual TCP test parameters, we have developed a web UI that allows the creation of templates for test measurements and this should help set the adequate parameters for tests for each host. The scheduling feature can select the template for the kind of test for individual hosts like the example of a test in the 10G -> 1G direction (screenshots in attachment for more details).

Regards,

Alex

On 16/07/2015, at 15:31, Matthew J Zekauskas <> wrote:

If you are talking about a TCP throughput test, and the MTUs are set correctly on the host interfaces, TCP should negotiate the lower MSS (max segment size) for transmission.  I don't believe it is user-specifiable on the individual tests.  Note that if there are tunnels, then the 1500 MTU may be larger than the biggest actual MTU of the path.

It could be worth limiting the window size to throttle the connection to 1GE in the 10->1 direction if there isn't enough buffer space to deal with a burst of data arriving at 10GE at the bottleneck.  (...or if you find that throughput is bad in that direction, then playing with the window size could confirm a buffer issue.)

--Matt, with a quick answer which others will no doubt elaborate on

On 7/16/15 2:10 PM, Alex Moura wrote:

Hello,

Quick question: is a case where one perfSONAR host does have 10GbE port with MTU set to 9000 bytes and the other side host does have 1GbE port set to 1500 bytes MTU, how one should set the parameters to run a throughput test on each side?

Thanks!
Alex