ExoSM paths TCP traffic inconsistencies

[lnevers@bbn.com]

Finding TCP traffic results are inconsistent for site to site connection set up by the ExoSM stitching between the GPO and Houston rack.

The following TCP performance was found from GPO EG to Houston EG (repeatable):

   1 TCP client = 20.4 Mbits/sec 
  5 TCP clients = 3.02 Mbits/sec
 10 TCP clients = 2.98 Mbits/sec

Why does throughput drop from 20 Mbits/sec as more client are added?

These are the results for the same requests in the opposite direction from Houston EG to GPO EG:

   1 TCP client = 7.95 Mbits/sec
  5 TCP clients = 8.81 Mbits/sec
 10 TCP clients = 8.66 Mbits/sec

Why such low throughput for an ExoSM VLAN?

The results above are inconsistent compared to other sites.

From GPO EG to UFL EG:                         
   1 TCP client = 27.5 Mbits/sec
  5 TCP clients = 139 Mbits/sec
 10 TCP clients = 278 Mbits/sec

From UFL EG to GPO EG:
   1 TCP client = 27.5 Mbits/sec
  5 TCP clients = 129 Mbits/sec
 10 TCP clients = 201 Mbits/sec

[ibaldin@renci.org]

Please be specific as to what bandwidth was actually set in the request?

[lnevers@bbn.com]

The bandwidth was not specified, iperf determines how much it can get. The actual iperf commands run:

1 Client:       iperf -c 172.16.2.2 -t 60
5 Clients:      iperf -c 172.16.2.2 -t 60 -P 5
10 Clients:     iperf -c 172.16.2.2 -t 60 -P 10

Each test ran for 60 seconds.

[lnevers@bbn.com]

The GPO to Houston tests were re-run and there no longer is any inconsistency in TCP traffic results. Closing ticket.