[Spread-users] Performance

Yair Amir yairamir at cnds.jhu.edu
Mon Jul 29 11:35:53 EDT 2002

Are all of your machines the same speed and all of them empty?

There are some tuning to make Spread run faster than the
out-of-the-box performance, because the out-of-the-box is designed to
work without many people contacting us. From your numbers, you have
substantial decrease, so you probably have another problem - probably
uneven machine power or too many losses on a machine, etc.

3.16.2 has no problem to support 100 receivers but not if each of them
is running on a different machine. For that you need a smarter design.
(or a more scalable version). I would also go with smaller messages in
case you have a problem with IPC queues (assuming you are using IPC
and not TCP for the local connections).

    :) Yair.

Jan> hello list!

Jan> we did some spflooder tests and we were increasing the number of hosts (linux).
Jan> we realized, that spread seams to get slower by each added host. 

Jan>         1.000 messages/100.000 bytes - time is averaged for all hosts 

Jan>         freiburg -> okinava

Jan>         15.831 ms (48,19 Mbps)
Jan>         15.251 ms (50,20 Mbps)
Jan>         15.269 ms (49,96 Mbps)

Jan>         freiburg -> okinava, galway

Jan>         17.898 ms (42,62 Mbps)
Jan>         18.123 ms (42,90 Mbps)
Jan>         18.482 ms (41,28 Mbps)

Jan>         freiburg -> okinava, galway, qosqo

Jan>         21.143 ms (36,80 Mbps)
Jan>         21.159 ms (36,50 Mbps)
Jan>         20.365 ms (37,46 Mbps)

Jan>         freiburg -> okinava, galway, qosqo, teguise

Jan>         25.692 ms (29,69 Mbps)
Jan>         25.663 ms (29,72 Mbps)
Jan>         25.075 ms (30,42 Mbps)

Jan> we are using spread-src-3.16.2.tar and our spread.conf looks like this:

Jan>         Spread_Segment {
Jan>              freiburg
Jan>              okinava
Jan>              galway  
Jan>              qosqo
Jan>              tequise
Jan>         }

Jan> but the main question is, how is it possible to increase spread's performace?
Jan> we will have just *1* sender and up to *100* receivers in our scenario.

Jan> and another question, does spread nead a lot of resources(cpu, memory)?

Jan> regards, jan

More information about the Spread-users mailing list