Thursday, June 14, 2007

Multi-Cores, SOA, SCA and the Incredible Lightness of Being

I received a link from a colleague regarding a recent article "SCA applies SOA principles to Apps Design" by Andy Dornan. The article starts off with a slant that SCA enables runtimes to take advantage of multi-core CPUs; further implying that while Moore's law still is applicable for making the processors themselves faster, that the real leverage in software gains comes from building applications that can have multi-core processors thrown at them. He theorizes that in order to really leverage the multi-core processor future, that applications must be broken up into small loosely-coupled components. And this was just the first paragraph.

Whoa. Where to start. I can say the article made me think, does SOA, or SCA for that matter really help applications deal with multi-core processors? Far be it for me to criticize semantic English, but multi-core processors imply same-box, and the focus of SOA and SCA relate to the wider, distributed network; which may be comprised of multi-core processor nodes. Certainly SOA does help application developers in arriving at application patterns and best practices that lend themselves to being advantaged by multi-core processors, or even more importantly a set of distributed processes and processors. Simply using container-based applications like JEE or even Spring put the majority of the burden of thread management in the hands of the middleware, where presumably there are appropriate controls and exploitation of multi-core processors have been meticulously built; that really doesn't have much to do with SOA or SCA. Billy Newport has blogged numerous times about applications and multi-core issues as well. While there has been lots of research in predictive instruction execution and pipelining, SCA offers more like a circuit board assembly than a flow chart (perhaps like BPEL) and its model doesn't really imply any ordering or sequencing which would afford any great advance in multi-tasking. Certainly SCA has asynchronous invocation concepts which would allow for a greater opportunity for muli-tasking, but I would hardly say they are transparent to the application.

To cut Andy a break, perhaps what he intends to say is the SOA and SCA programming style of providing declarative dependencies and policies allow the runtimes to better understand the application and therefore, de facto have a better chance of exploiting the hardware available to solve the problem.

The article then diverges into other topic areas for which I also have some comments.
  • Java is certainly an important language for SCA to embrace and to treat in a first class manner, but it is hardly the only language. Certainly at least BEA and IBM and keenly interested in SCA assemblies with BPEL implementations, but there is lots of activity in other languages. Of course WebSphere is going to focus on Java -- it is a Java runtime. What will be interesting is how vendors choose to slice and dice the type of component implementations they support and in what form factor they will deliver that support.
  • I didn't quite understand his comments about requiring a higher-level intermediary to embrace WCF components. Certainly WCF components can consume or be exposed as Web Services, using WSDL and XML for interface definitions -- SCA embraces both service consumers and providers. Perhaps I'm missing some other point Andy is trying to make. Certainly, as I've blogged about before, SCA embraces existing (legacy) technology and I don't understand why anything would have to be rewritten. This seems just wrong to me.
  • I don't agree with the statement that the vendors believe SCA is needed to achieve "tru(e) platform independence." What SCA allows you to do is to compose and assemble applications written in a variety of implementation languages, hosted on a variety of platforms and insulates the business logic developer from having to have unnecessary knowledge of these implementations.
  • Microsoft is not interested in application portability, so certainly aspects of SCA are not interesting to them; however, the author does seem to be unaware that SCA, albeit in its original proprietary, has been shipping in WebSphere Process Server 6.0 for some time now.
  • Service Data Object is not an optional subset of SCA. SDO is orthogonal to SCA, but is rather complementary... to be honest, I just don't know what Andy was trying to articulate in his closing paragraph about SDO.
At the end of the day, I suppose its nice to see some articles coming out regarding SCA.... even if the ideas in the articles, I can't get my head around.

1 comments:

Terry said...

I think your last point is the most interesting -- it's nice to see articles being written about SCA. And let's face it, although SCA is pretty simple from a programming language standpoint (purposefully), it'll take time for people in the industry to get their heads around SCA and SOA and what folks should really be using it for. I suspect this won't be the last SCA article where the author gets some fairly fundamental things wrong... such is progress.