Twenty million flies ...

[Mobiwan]

I had a dispute with a colleague "from the SOA camp" this morning on using XML based messaging to couple distributed systems. The discussion was ignited by a performance issue he is currently facing with two systems suffering from serious latency while exchanging XML over the Internet.

My comment on that binary protocols such as Ice (or CORBA or even RMI for that matter) are superior to XML based protocols with respect to performance generated complete uproar: RMI and CORBA would generate tightly coupled systems, binary protocols were not state-of-the-art, RMI and CORBA were out-dated pre-paradigm-change concepts no one would use today, no SOA implementation is using binary protocols and, ta-daaa, the inevitable "Why would all others do it if it was wrong?"

These "arguments" really can make your blood pressure rise. I did my best to explain that coupling has nothing to do with using XML and that level of adoption has nothing to do with quality didn't really move him. I think the incident shows two serious problems Ice is facing: FUD and the old 'all-others-are-doing-it-so-it-can't-be-wrong' argument.

Unfortunately I didn't have papers or demos at hand to support my position which leads me to the core message of this posting:

It seems as if we need more argumentative support especially for the enterprise application community:

* through easy to grasp performance comparison,

* through demos showing how easy it is to use Ice as an alternative communication layer for the average application server hosted app,

* by showing people how the layers of an enterprise application can interact more efficiently when using Ice,

* by explicitly showing the steps necessary to substitute SOAP messaging with Ice messaging in an enterprise application,

* by showing how to build complete SOA apps using Ice.


If we don't want to stay 'niche' with Ice we'll have to infiltrate the mass market. And mass market means enerprise applications and SOA these days. Demos and arguments should more directly target those areas!

With it's current packaging, to the uninitiated Ice looks like those parts of application development, most developers don't want to deal with these days, that is, low level communications. And the demos delivered with it enforce this impression, I think.

It is no longer enough to have the 'Hello World' demo and leave the rest to potential customer's imagination. People don't have the time to get a grasp with technology, they want proven solutions or at least examples directly mimicking their environment.

What do you think?

Cheers,
Mobi.

[kwaclaw]

Originally Posted by Mobiwan

I had a dispute with a colleague "from the SOA camp" this morning on using XML based messaging to couple distributed systems. The discussion was ignited by a performance issue he is currently facing with two systems suffering from serious latency while exchanging XML over the Internet.

My comment on that binary protocols such as Ice (or CORBA or even RMI for that matter) are superior to XML based protocols with respect to performance generated complete uproar: RMI and CORBA would generate tightly coupled systems, binary protocols were not state-of-the-art, RMI and CORBA were out-dated pre-paradigm-change concepts no one would use today, no SOA implementation is using binary protocols and, ta-daaa, the inevitable "Why would all others do it if it was wrong?"

These "arguments" really can make your blood pressure rise. I did my best to explain that coupling has nothing to do with using XML and that level of adoption has nothing to do with quality didn't really move him. I think the incident shows two serious problems Ice is facing: FUD and the old 'all-others-are-doing-it-so-it-can't-be-wrong' argument.

I completely agree.
As someone interested in software architecture I have yet to see anything about SOA that is actually an innovation.

I think what would help is a white paper titled "SOA with ICE", maybe coupled
with a small demo application. This would help even though SOA is just a buzz word.

And secondly, the "tight coupling" argument has to be addressed specifically. I have read part of a recent blog discussion involving Michi Henning and others about that very issue, as well as earlier postings on the xml-dev mailing list, and even some part of the famous dissertation paper about the REST architecture. I believe one of the central points of the REST and the SOAP camps is that it is better to be able to receive a message and do partial processing (based on how much of it you understand) than to not receive it at all because the type system does not allow it.

So, one has to address the perils of partially understood messages.
One also has to address that tight coupling has nothing to do with type safety, unless one considers a common understanding of every part of a message as tight coupling.

And maybe we have to understand why it is possible that some large systems are using web services and still seem to scale. Is it because they are throwing so much hardware and bandwidth at it? In that case one would have to demonstrate the savings achievable with ICE (or even CORBA).

Karl

[stephan]

Dear ZeroC people,

As can be seen in our signature (and as you know ), we have been writing two articles on Ice together which demonstrate the ease of using Ice.
The bottomline of those article is that Ice definitely is competetive against todays in-fashion technologies.

We have, however, only shown this at the implementation level, not at a
more generic, and probably business-people understandable, level.

I think what the other Stefan was referring to is that a demonstration on a higher level would be more interesting. Imho, this could be an average-decision-maker understandable display that Ice can easily be used as the basis of a Service Oriented Architecture, with all the advantages and features that make people go mad nowadays.

It would be very interesting to hear what you guys think about the idea and what your impression about the acceptance of Ice in the industry is...