incubator-s4-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Matthieu Morel <>
Subject Re: Performance of the stream between adaptor and PE
Date Mon, 26 Nov 2012 12:12:45 GMT

I assume you are using the 0.5 release. It has known performance culprits, in particular for
remote communication. That's why one of the focus of 0.6 is actually performance.

Latest development in branch S4-95 provide more than 250 000 *remote* messages per second,
but we still need to review the latest changes before integrating. (very soon).

So: yes you might face relatively low performance with the 0.5 release due to communication
overhead, but forthcoming updates will definitely change the situation, with minimal changes
to the API.  (actually, probably the only change is the option to specify the parallelism
of streams). You might even try the S4-95 branch if you want a preview now.



On Nov 26, 2012, at 12:52 PM, German Blanco wrote:

> Hello,
> I am doing some very basic performance tests and I found something strange.
> My configuration has two clusters, one with an application and one with the adapter.
> The application has two different PE.
> It seems that if I launch a loop just creating events and sending them between the two
PE, the throughput is of a few hundred of thousands of events per second. However the same
test between the adaptor and one of the PE has a throughput of just above ten thousand events
per second.
> I guess the reason could be the inter-cluster communication, but isn't the performance
drop too much?
> Anybody has had similar results? Any ideas of what could be wrong?
> Regards,
> German Blanco

View raw message