integration ecosystem

« Greetings From Anaheim-SHARE 2005 Main SHARE - Day 2 »
February 28, 2005
SHARE - Day 1
There are two tracks I'm trying to follow at SHARE, first is a WebSphere track, which appears heavy on the MQSeries, and second is the web services/SOA track.
I saw a few things today, and am halfway relieved to find that I'm not the only one confused by SOA and ESB, as apparantly IBM is too!
I think there is a big problem with Enterprise Application Integration (EAI). There are too many ways to skin the cat. Not that options are bad, but it seems awfully hard to hang one's hat on anything these days.
I recall the days not too many years ago when we went from 'web services will solve everything', to the retrenchment, where it was 'web services won't solve everything, and when you can drop down to more performant techniques such as IIOP, you do that'. We'll, were back to 'web services will solve everything'.
Actually, it's 'web services and SOA will solve everything, and we have a BPEL engine, but on the other hand here's the WebSphere MQ Broker architecture,which is based on messaging'. They are both message-based workflows, assembled out of nodes of various types into a functioning application.
Web services really doesn't dictate SOAP over HTTP, JMS type queues can also be the transport. WSDL itself is being beefed up to be able to describe all sorts of endpoints. At least on the surface, could you not express your message broker flows as BPEL, and describe the services invoked by the nodes as WSDL, and use SOAP and XML to represent the messages and data internally? MQ Broker uses the idea of a parsed tree to represent data internally, and it's only marshalled into a wire format after it leaves the broker...sounds like an XML DOM tree! Maybe the answer is glaringly obvious to others.
I see the same thing with Oracle software, where apparantly the emphasis has shifted from things like processConnect to things like BPEL process manager. Perhaps it's just an inflection point where hub-and-spoke messaging/transformation architectures are giving way to web services/BPEL SOA architectures. Certainly, the problem space they are all trying to address is quite related.
Now throw in the Enterprise Service Bus. Here is an attempt at a spec for loosly coupled interfaces, described by WSDL, applied to a lightweight container model, based on messaging at its core, orchestrated, with ideas like itinerary-based routing, and using SOAP and WSDL to abstract the messages from their underlying protocol. I've been simultaniously looking at JSR-208, which describes an actual architecture for an ESB, and constrasting it with the IBM view of an ESB. IBM seems to consider an ESB to be a 'pattern' that can be expressed through the WebSphere product line, leaning heavily on WebSphere MQ. I found the JSR-208 specification dismissed out-of-hand by some of the WebSphere folks. IBM is glaringly not on the JSR-208 Expert Group. I don't know the details, philosophical, technical, and market/product positioning, that bring about this seeming departure, but will be interested to find out.
Look, I don't know the right answer. I'm using this week to try and make my way through some of this. I hit the floor tonight to talk to some of the IBM types and see some of the various products, but everyone looked pretty burned out, so I'll hit-em fresh in the morning.
It looked like CICS was staging a bit of a comeback. I think a reorg has happened that pulls CICS out from the WebSphere cloud into its own space. I wonder if there was a reaction by IBM customers to seeing WebSphere on everything like there apparantly was against Microsoft for having a .Net on everything. I also think folks are realizing that it doesn't pay to throw away millions of lines of code. It was mentioned that Gartner had come up with a figure of $25 per line of code in an enterprise, considering all costs of writing and maintaining an application. A million lines of code becomes quite an investment. If stuff is working, only an idiot would throw it away, right! I think the tone is to extend the investment in CICS by opening up the code using SOAP, good old messaging, and the CICS Transaction Gateway/JCA. I saw business examples using each methodology, and it's high on my list to get a good answer on the relative merits of each, and whether one technique is emerging as the primary vehicle. I'll bug the IBM guys on the floor on this one.
And how many folks are running WebSphere on Z/OS? It seems like you don't run into lots of these folks, but that's a most inaccurate poll. I sensed a bit of schizophrenia around WebSphere and CICS. I thought they were converging, and CICS was going to look like an Abrams tank of a J2EE platform, but now it seems more like a 'back-tier' complement to the 'mid-tier' J2EE platform. That may have been it all along, I have not followed CICS religiously since leaving the Cobol/CICS/IDMS game in earnest.
I have spent most of my time looking at the Oracle app server architecture, so I was happy see some info on the latest WebSphere architecture ideas. I'm planning on putting this up in the CoLab when I get back to look in detail, but the broad strokes of cell, node, and app server look an awful lot like farm, cluster, and instance. That's a surface impression, I'll admit, but I'll be getting into the nitty-gritty soon enough, and at least their general architecture doesn't look like some big departure.
So, SHARE seems to be stacking up like this:
SOA versus MQ BrokerESB a-la JSR-208 versus ESB as a 'pattern'CICS moving forward versus the WebSphere J2EE application serverConnecting to CICS...SOAP, Gateway, MQ?
I hope to get some level of clarity, though it could also be my jet-lagged head. SHARE seems like a good show, and almost like a supplement to JavaOne last year...I wish IBM would come back to that, as I feel like I'm only seeing the whole story now.
Nuff for now...more questions then answers, of course. BTW, this was a pretty good primer on JBI and JSR-208, plus you can kick back and watch it.
Posted by conwaym at February 28, 2005 09:38 PM
Comments
Mike, good update! It's interesting too that IBM is pushing PHP now with the integration of IBM’s Cloudscape database and Zend’s PHP environment (Zend Core for IBM) [InfoWorld: an “out-of-the-box” PHP development and production environment allowing corporate users to more easily develop, deploy, and manage open source applications.] I'm curious if folks are talking about that at all.
Posted by: pbm at March 1, 2005 08:57 AM
If I see IBM pushing anything at SHARE, it's SOA and their messaging middleware!
Really, I have not heard a whiff about PHP here, it's really hard-core enterprise types that want the skinny on the latest DFH-whatsis within CICS.
The stuff at SHARE is very 'conventional' in the sense that it is very enterprise and integration oriented. That's both good, and bad. I felt like IBM pretty much said 'you JBoss kiddies go off and play with your toys, we're big kids and we have bigger toys'. On the other hand, this place if full of lots of folks doing serious business, moving serious transactions and $$$ using technology that the OSS types claim doesn't work or has 'gone away'. I saw that sillyness too with all the NFJS folks dissing JavaOne, when I found both NFJS and JavaOne quite valuable.
I've programmed in PHP, and it seems good for some things, but I don't really see anything earth-shattering there. I say good-o for PHP!