The State of SCA: An Update
Friday, May 09, 2008
I moderated a panel on Service Component Architecture (SCA)
at JavaOne last week. I was also the moderator for last year’s SCA panel, and several of the same people were on the panel with me this time. While the things we talked about were broadly similar, two things stand out about what's changed in a year.
The first is that SCA is real, or at least part of it is. One of the things the SCA specs define is an XML-based language called the Service Component Definition Language (SCDL). SCDL is meant to provide a vendor-neutral way to describe how components created in various technologies, such as Java, BPEL, and Spring, are configured and wired together to create applications. Vendors were showing SCDL in real products on the JavaOne floor—Oracle had an especially nice demo—and so it's clear that this part of SCA is seeing some success.
Whether SCDL will in fact provide much cross-vendor portability remains to be seen. As usual, this depends on how many proprietary extensions vendors add. Still, a standard language for describing the components and assembly of an application is a useful idea, and the signs so far are promising.
The second thing that stands out after a year is less promising: It’s the confusion around how to write SCA components. Along with SCDL, the SCA specs define how to create components using several different technologies. Yet the various SCA vendors and open source projects can’t agree on which of these to implement. SCA support for Spring components, for example, is hit or miss: some SCA offerings support it, some don’t. BPEL is much the same—Oracle is a big fan, while the open source Fabric3
currently has no BPEL support.
And just as it was a year ago, support for SCA’s new programming model for creating Java components is uneven. As I've written before
, I believe that this aspect of the spec is really important--it unifies the diverse approaches of Java EE much as Microsoft's Windows Communication Foundation (WCF) unified the diverse programming models in the original .NET Framework. Yet this part of the SCA standard has always been contentious. At last year's SCA panel
, for example, the SAP rep asked the audience who wants to see a new programming model for Java components and (unsurprisingly) got no hands raised. Accordingly, SAP has been a leader in defining an alternative way to create Java SCA components as EJB 3.0 session beans. This alternative is a superset of SCA's original component model, so it's not a wholly new thing. Still, the challenge for developers and decision makers is to choose among these various options, and so creating more of them is problematic.
Some existing SCA offerings, such as Fabric3, implement the original Java programming model for SCA components and apparently have no intention of supporting the EJB-based approach. SAP, by contrast, explicitly told me that they have no plans to support the original Java programming model; they're going with the EJB-based approach. IBM, ever the big tent, is supporting both.
The stated goal of SCA is to provide application portability. Widespread support for SCDL is an essential part of this, but so is agreeing on how to create SCA components. For SCA to really improve portability, the vendors and open source projects that support it need to agree on how their customers should create components. If they don’t, SCA risks becoming yet another standard that doesn’t provide much benefit to the people who use it.