Re: Comparision with TIBCO, Vitria and MQ Series?

Hi.

The event rate is being measured in very different ways. Some publish/subscribe middleware, such as Vitria, are based on a "hub-and-spoke" architecture. This means an event is always sent to a "channel server" (the hub) first, and from there it is delivered to each subscriber separately, by using one TCP/IP message per subscriber.

Now some vendors of hub&spoke middleware measure each published event actually as n+1 events being delivered! Hence one should ask for throughput numbers where the "number of published DISTINCT events" is measured.

I'd guess (just my personal opinion) that the TIB performance shown below was measured the "right" way, but that the Vitria number was not. It always takes about one millisecind to send an event via TCP, unless a) either aggressive buffering is used at the sender, or b) all the events delivered to the subscribers are added up.

I might be wrong, but as a matter of experience I'm a bit suspicious when somebody claims thousands of events/second being sent through a p&s channel.

Besides that, TIB (and iBus) use IP multicast communication and thus use network bandwidth much more efficiently than Vitria.

On this line, the iBus publish/subscribe middleware (http://www.softwired-inc.com/ibus) achieves about 400 events/sec (I.e., DISTINCT events published).

Silvano

In article <0685A427A719D11197BB00A024D399450BE581BC_at_delphi.sapient.com>,   John Harris <jharri_at_sapient.com> wrote:
> Hi Jonathan,
>
> I am currently involved in work on enterprise integration projects. I
> would be interested in comparisons of TIBCO, Vitria and MQ series
> against Oracle AQ. Or comparisons of any other message based middleware.
> I believe Oracle would probably come in last. Has anyone got any info?
>
> Vitria publish some staggering performance metrics for published
> events/second. I.E. sustained message rates of 3000 events per second
> with a message size of 100 bytes over 12 channels.
>
> They claim on the same hardware That TIBCO (TIB/Rendezvous) could only
> handle about 200 events per second
>
> My wild guess, with no evidence at all, is that Oracle would be way
> below these numbers (10-100)
>
> > -----Original Message-----
> > From: Jonathan Lewis [SMTP:jonathan_at_jlcomp.demon.co.uk]
> > Posted At: Thursday, January 07, 1999 11:05 AM
> > Posted To: server
> > Conversation: AQ in Oracle 8
> > Subject: AQ in Oracle 8
> >
> >
> > Has anyone stress-tested AQ in Oracle 8 ?
> >
> > Any ideas on how many actions per second you could
> > get from it if you used AQ as a way of dispatching
> > some form of audit trail which recorded in an audit
> > database the actions taking place in another database ?
> >
> > (My guess is - not very many - but I'd like to find out the lazy way)
> >
> > --
> > Jonathan Lewis
> >
> >
>
>

-----------== Posted via Deja News, The Discussion Network ==---------- http://www.dejanews.com/ Search, Read, Discuss, or Start Your Own Received on Sat Jan 30 1999 - 10:55:20 CST