[adrian.brock]

We should write a MockSocketFactory that simulates:

* latency
* bandwidth saturation
* backlogging
* random broken connections
* network splits
* etc.

:-)

Probably something that belongs in the remoting testsuite.

[adrian.brock]

Also, a lookback device is usually multi-threaded.
A real NIC is not. TCP/IP just makes it seem like it is.

[timfox]

Note that if you're processes are on the same machine, and you're using TCP to send messages, your NIC will *not* be used, so performance measurements will ignore the network and bypass it using TCP's loopback facility.
I suggest having the 2 processes on separate machines to really measure the throughput and/or message rate.

The perf framework tests with multiple clients on multiple boxes. Now problems with JBossSerialization are resolved I can get results for this.

As mentioned, the current results are (knowingly) obtained using the loopback.

I've been testing in 3 main configs:

1) Client and server in vm
2) Client and server same machine different vms (using loopback)
3) Clients and server different machines.

[belaban]

Okay, great. Just wanted to make sure you are aware of the (incorrect) performance gain using loopback. And as I can see, you are.
Thanks,