Ten things you need to know about SOA
SOA (service oriented architecture) is a big deal, I like it. But it isn't the be all and end all of computing. Here are ten things you need to know about SOA.
1. You can't sell SOA. SOA allows your company to be more flexible. SOA enables the agile enterprise. Oh, yes. But you can't build a business case or cost justification out of flexibility or agility, you can only build that based on resolving a business issue or issues. In appropriate circumstances SOA may enable a business solution to a business problem: that's all.
2. Even if you could sell SOA you couldn't because you can't describe to a business person what it is. The truth is that there is no good definition of what SOA is. Even as a concept it is fragile and different vendors and analysts will give you different (and nebulous) definitions. If even the IT industry cannot agree on a definition, how can you expect the business to understand it? The best that can be said is that it represents a set of enabling technologies.
3. Business Process Management (BPM) is not SOA. You can do either without the other though doing SOA without BPM might be tricky. 4. BPM engines represent a potential bottleneck for SOA. If everything is built around BPM and your services have to keep referring back to the BPM engine for instructions then that engine can become a bottleneck. You may therefore need to have multiple such engines with an “engine of engines” for orchestration. A better idea might be to have intelligent services, where appropriate, which understand their own routing and can maintain state information: thereby reducing the calls necessary to the engine.
5. BPM isn't enough anyway. It's fine for fairly simple processes but can become impractical when environments are very complex and, especially, where the business is exception-driven. This is the role of (complex) event processing (CEP).
6. Most vendors in the SOA space acknowledge the potential role of event processing but don't understand it. For example, I have seen SOA presentations where CEP is confused with operational BI-style event processing. Certainly there is a place for this in SOA (throughput monitoring for example rather than the instance monitoring of BAM [business activity monitoring]—though the two should really be combined). Oracle, which does understand event processing, has CEP at level 5 in its maturity model for SOA: which is fine, except that CEP can be implemented totally independently of SOA.
7. You don't need to use SOAP (simple object access protocol). Funnily enough, this isn't as simple as it is supposed to be—there are alternatives that are simpler. 8. One of the biggest potential paybacks for SOA is in the ability to reuse services. But how do you make this happen? We couldn't do it for objects and we couldn't do it for components so why do we think we can do it for services? We can set up SOA governance and implement IT policies and rules but does that mean that developers will abide by them? When the pressure is on? When this job has to be finished by tomorrow? 9. And talking about governance how is this going to work—what is the relationship between SOA and data governance, for example? If part of the purpose of governance is to establish the ownership of processes and data then that is a problem, because ownership implies responsibility and people will shy away from the latter if given half a chance. The theoretical models put about for governance are all very well but if they can't be applied in the real world (at least some of the time) then we need more pragmatic sets of “the best we can reasonably get” practices rather than aiming for the ideal all of the time.
10. Data is largely ignored (IBM is an honourable exception here) by most of those talking about SOA but if SOA is about undoing the Gordian knot of spaghetti that most company application architectures look like then shouldn't the same be the case for the similarly complex data environment?
There a probably a lot more things that people need to know about SOA but I have run out of space. Perhaps readers would like to add some more points?
Copyright © 2006, IT-Analysis.com
Sponsored: Global DDoS threat landscape report