Mainframe SOA – an oxymoron or the bees knees?

Over the last 15 months or so IBM has been putting considerable resource into making the mainframe (zSeries) SOA ready. The main thrust of this process has been to move all the z infrastructure products to a common code base for the distributed and z products, whilst still enabling them to take advantage of the unique features of z. This process started with DB2 V8 and WebSphere Application Server and the first phase will end with the delivery of WebSphere Process Server for z/OS and WebSphere ESB for z/OS in the middle of 2006.

IBM has been working with Siebel to implement Siebel’s distributed applications on z. The results have been impressive; it really is possible to take an application developed for a distributed platform and drop it on to the mainframe. It works without significant change and performs very well. After the move the application inherits all the quality of service benefits of z, including reliability (five 9s now being the minimum expected), security and scalability.

So having shown it is not an oxymoron do people believe it is the bees’ knees? There are indications in the affirmative. IBM has already worked with ten other leading application vendors to enable them to move and is working with twenty more. These vendors would not invest in this way unless they could see a significant return on investment. It appears they can see it because the cost of the move has been dramatically reduced and some of their large customers are asking for, or demanding, z versions of the applications.

The customers are asking for z versions, in many cases, because they have been affected by security violations or reliability outages that have impacted the business reputation and/or the bottom line. In other cases they would like a z version because they recognise that there is a major risk of such an event occurring and they would like to avoid it.

What I have talked about so far is moving applications and the services they provide to z, but what about the rest of the SOA infrastructure, especially the business process management (BPM) and enterprise service bus (ESB) layers?

I have been learning about how Rabbis through the ages reach conclusion about biblical regulations. They have a set of seven rules, the first of which is ‘a fortiori’ which says that if a statement is true about something ‘light’ then it applies ‘all the more so’ to something ‘heavy’. So in my view applications and their services are ‘light’ in comparison to business processes, because the whole business process is more critical to the success of the enterprise than the individual services. If the process fails it really does not matter if some of the services are still working.

So the Rabbis would conclude that if the mainframe is right for the applications and their services, then it is ‘all the more so’ true that the mainframe is the right place for the business processes and the ESB and the Business Process Server that support them. This suggests that even if a company does not move a working application to the mainframe (it ain’t broke so do not move it) they should implement the new business processes on the mainframe.

My recommendation is that companies that are serious about SOA, and do not have a religious aversion to z, should be planning to implement their business processes on the mainframe. To enable this, development should start on the distributed version of the software with a plan to install the production version on z when the software becomes available in 2006.

I only occasionally make predictions but I will stick my neck out this time, and be more bullish than IBM, by suggesting that by the end of 2007 Business Process Server for z/OS will be installed in more than 50% of the z/OS customers.