原文

Opinion: Would it be better to move from SOA to SOI?

Posted by: Milton Smith on February 18, 2005 @ 09:36 PM

SOA is great but it’s a lofty concept. I’m looking forward to the day when SOA becomes SOI, Services Oriented Implementation. I imagine an application server where I can run Java, C++, maybe Perl or other popular languages. All managed under one engine. The engine would adhere to specifications driven by the world’s foremost experts. I need to protect my investment, of course.

Today’s application servers are good role models. They provide basic functionality for: thread and database connection pooling, application deployment, clustering, web based management, and more. It’s been tremendously handy keeping all these details out of our code. Remember the day when we had to code all the details as well �C ouch. Moving applications from one application server to another is relatively easily. Granted, not without a little pain and frustration, as us salty dogs know.

Up to this point our visionary SOI server can run Java, C++, and Perl applications. All managed from a single application server. Wow, a Herculean dream, but it’s not over yet. We are still missing: diagnostics, security model, integration framework, property management, auditing, license management, analytics & instrumentation, persistence, and much more.

Sadly, today there are no SOI servers. Businesses are solving their own problems using a mix of existing technologies and custom code. Effectively is analogous to each business writing their own application server for servlets; we would all agree this is craziness. We all need the same features but attack the problem differently. Of course, anyone can argue each company has different needs. How could implementations account for this? Well, the industry did it with servlets! The key is to separate the service component(e.g, property management) framework from the implementation. In the case of property management imagine a Window’s program being able to leverage a Java property or the other way around. Most of us could build such a system but it would be a one off solution. And solutions not build upon standards are not information assets that can be transferred easily.

Think about the problems this can solve! I believe with the right talent, seeding and buy-in from vendors, we could get there. There is already work on a number of fronts, both technology and specification:
Diagnostics �C log4j, JMX, and SNMP
Security model �C Liberty Alliance, SAML
Auditing �C log4j, Sarbans-Oxley
Property Management �C JNDI, Preferences, JDO, ORM, etc.
License Management �C only vendor technologies. Little in the way of adopted standards or open implementations.
Persistence �C RDBM, JDO, ORM, the list goes on
Integration �C MOM, ESB, SOAP and Web Services

Is this a reasonable dream? Can all this be stuck under the hood of one car? Is the enterprise destine to be a permanent mix of technology and competing standards? Before you answer, think about how many dollars are spent when one company acquires another and both must integrate. Or a service provider paradigm, where one vendor must integrate with another to provide a service. Multiple this times the number of occurrences in the country. The cost is huge! Is anyone perusing such challenges or interested in doing so?

Milton Smith
[email protected]