The Faban master normally does not need tuning. You should be able to just drop it in and run. The only exception is for very large test rigs with very large number of systems - we talk about 50 or above - and each logging a lot of information to the log server in the Faban master, you may want to make sure the log server does not become a bottleneck.
The log server tunables are the last part of the harness.xml file, shown below:
<logServer> <!-- Don't touch this unless you really know what you're doing -->
<port>9999</port>
<listenerThreads>1</listenerThreads>
<listenQueueSize>-1</listenQueueSize> <!-- system default -->
<serviceThreads>
<core>2</core>
<max>10</max>
<timeOut>300</timeOut>
</serviceThreads>
<bufferSize>2048</bufferSize>
</logServer>
Please do not change the port configurations in harness.xml. The consequences are widespread far beyond the Faban harness. Other than that, the log server has a classic web server architecture with usually one listener thread and multiple service threads. The listen queue size specifies the size of the queue maintained by the underlying operating system. We should leave it as the system default unless we know for sure that setting it otherwise will improve performance. The listener threads beyond 1 should work but have not been well tested.
So we are down to the two common tunables, the service threads and the receive buffer size. The service threads are set to 2 by default and can increase up to 10 under load. You may want to increase the max number of service threads if logging is from many logger nodes and the amount of messages is very high. The receive buffer size of 2048 bytes is usually adequate. You can obtain better performance by increasing the size if the log messages are generally very big so that one message does not incur multiple I/Os.
Other than these few tunables, the Faban master should never need any other tunings. Good luck and have fun with it.