[c-nsp] Weird throughput issue

[Lukas Tribus]

> UDP tests to the site are ok as we can force 85-90M and the
> site receives it with very little packet loss. 

Packet loss none the less. If you stay (well) below the CIR you are not supposed to see any packet loss. Also, check for jitter and OOO packets. All those things matter, and will make TCP behave badly.

For example, if a single TCP session is load balanced through different queues or paths on the provider network, jitter and OOO delivery will occur killing TCP performance like you describe while the RFC2544 reports wont report anything unusual.


Lukas