[Spread-users] Performance related question

Henrique Teófilo henriquetft at gmail.com
Fri Jul 10 23:11:29 EDT 2009


Hi folks,

I would like to ask some questions about Spread performance, because I want
to be sure that Spread toolkit is running at its best.
I have two Java applications which were implemented using Spread and JGroups
toolkit. (the spread implementation does not use the Java API but a
implementation that makes native calls through JNI)

In one application I have from 2 to 6 members in a group, where all of them
multicasts totally ordered messages(agreed service). Each one of the
proccesses sends messages one after another in a fashion like: for (i = 0; i
< 1000; i++) { multicast }. With small messages (up to 100 bytes) I got very
good performance, which easily beats the application implemented with
JGroups toolkit. However, with 1000 bytes messages the performance falls
down quickly, and I got spread slower than JGroups with 10000 bytes
messages. Is there some optimization to do in this case? Something to do
with fragmentation of messages??

The other application is pretty much like the first one, but it is a
simulation of remote procedure calls, which multicasts one message and waits
for the response before going on. In this case, I have worst performance
even with small messages. Is there some optimization to do in this case??
The spread toolkit delays the sending of a small message to do something
like TCP nagle's algorithm?

Thanks in advance,

Henrique Teófilo
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.spread.org/pipermail/spread-users/attachments/20090711/6c423656/attachment.html 


More information about the Spread-users mailing list