Re: Advertisement Discovery Performance

Mike Cumings <[email protected]> Fri, 9 Sep 2005 07:55:19 -0700
Newsgroups gmane.comp.java.jxta.user
Message-ID <[email protected]>
Hi Geoff,

How are you configuring your JXTA net with regard to rendezvous?
Does your application have the peers auto-promote themselves
to rendezvous after some period of time, or do you have separate
"infrstructure" peers that your grid peers are using?

The reason why this comes into play is that discovery takes place
via rdv peers.  The default time for an "edge" peer to promote
itself to a rdv peer is 5 minutes, matching your application observations
quite closely.  See RendezVousService.setAutoStart() for a good
starting point.

HTHs,

Mike

On 9/9/05, GEOFFREY DIX <[email protected]> wrote:
> Hello,
> 
> I am currently trying to run performance tests on an application I have
> written using the Java JXTA implementation (v 2.3.x).  I begin by starting
> up 12 peers on different nodes of a linux cluster.  Each peer creates and
> advertises a bidi pipe and all peers are members of the same group.  I
> discover the pipe advertisements by name (all advertisement names are built
> using <PeerGroupName>:<PeerName>) by making a call to getRemoteAdvertisements
> using 15 as the number of advertisements I would like to retrieve.  I wait
> about 10 seconds before sending the next getRemoteAdvertisements request (I
> have tried many different combinations - 1sec, 30secs, 60 secs) to give my
> discovery listener time to parse the advertisements and attempt to connect
> to the pipes.  Currently it takes about 5-10 minutes for all the peers to
> discover and connect to each other peer (though sometimes the discovery does
> not succeed for all peers).  My problem (finally) is that since I am still
> in the process of making changes to my software baseline I would like to
> find a way to improve the speed at which peers are discovered and wonder
> if there is a more efficient method for discovery that I am not aware of?
> 
> Thanks in advance for any help/suggestions you can offer.
> 
> Geoff
> 
> 
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: [email protected]
> For additional commands, e-mail: [email protected]
> 
> 


-- 
Mike Cumings