Re: xmlBlaster messages per second with many connected clients

[Marcel Ruff <ruff at swand.lake.de>]

Kyle Cordes wrote:
> 
> > > With CORBA, 435 messages/sec are delivered (taking ~23 secs to
> deliver a
> > > publish to all 10,000 clients)
> 
> > Yes this one is correct.
> 
> Assuming that the COBRA implementation is connecting to itself via TCPIP
> (not some kind of local shortcut),

No shortcut here.

> it looks like you are not seeing a
> large degradation in total throughput as the number of connect clients
> grows... good!  Your infrastructure (CORBA server) must be doing a good
> job of keeping lists of clients, lists of outgoing messages, etc. in
> data structure that don't degrade badly (no linear searches...).

Honours to Mr. Brose and his team, the JacORB maintainer.
(i've forwarded this 'JacORB good news' message to the jacorb list as
well
this message is part of a current performance discussion on
xmlBlaster.org)

> 
> By building on top of COBRA (as oppossed to your own wire protocol) a
> lot of these issues are dealt with by someone else.
> 
> Any idea what the bottleneck is?  In this test, was XML being parsed,
> etc., or just parsed once and send to 10000 clients (most likely)?

Yes, just parsed once.

When a message for a topic arrives the first time, the xmlKey and the
xmlQos
is parsed (one DOM and one SAX parse).

Additionally the xmlKey with topic meta data is merged into
a big DOM tree, allowing future XPath queries.

If subsequent messages from the same topic (same key oid) are arriving
the two parse steps are done as well, but no merge is necessary any
more.


Marcel

> 
> [ Kyle Cordes * kyle at kylecordes.com * www.kylecordes.com ]
> [ Training and Development Services: Java, Delphi, PHP,  ]
> [ ASP, ASTA, Web Applications, n-tier systems, etc.      ]
> [ Delphi developers: Visit the BDE Alternatives Guide    ]

-- 
Marcel Ruff
mailto:ruff at swand.lake.de
http://www.lake.de/home/lake/swand/
http://www.xmlBlaster.org