[Spread-users] Two questioins

[Jonathan Stanton]

On Tue, Mar 19, 2002 at 03:17:03PM -0500, Huaigu WU wrote:
> Thank you for your answer!
> But I am testing the performance of group communication, so I need to send
> messages faster and faster. in this case, whether are there some
> machinisms that can adjust the frequency of receiver in order to make
> receiver receive faster messages?

Certainly, but, just to be clear, the performance of the receiver is the
performance of YOUR client application that uses the spread api, not the
perforamnce of Spread itself. 

It also depends on what aspect of performance you are trying to evalutate.
If you are measureing throughput when there are multiple senders, just try
running daemosn on a few machines and run the spflooder program that we
include with Spread and you should achieve close to the wirespeed of the lan
(once message overhead is subtracted off). The spflooder program sends
messages as fast as possible, WITHOUT getting too far ahead of the receiver.

If by faster, you mean lower latency, then throughput and lots of messages are
not really needed and you should just be able to send them at whatever rate is
sustainable. 

The bottom line is if you want to measure Spread's performance, you do not
want your applciation to be the bottleneck (which is what happens if you get
disconnected, Spread was running faster then your app), you want to organize
your app so that Spread becomes the bottleneck. This might require using
multiple processes or machines as receivers with each one only handling part
of the traffice, while the daemons have to process all the traffic for all of
the clients.

I hope this helps, as I'm not quite sure what you are trying to do.

Jonathan

-- 
-------------------------------------------------------
Jonathan R. Stanton         jonathan at cs.jhu.edu
Dept. of Computer Science   
Johns Hopkins University    
-------------------------------------------------------