Last week we attended a workshop that brought together people from universities across the UK who are working on various forms of Service-Oriented Architecture (SOA) to underpin their business IT. The workshop was organised by Nikki Rogers at the University of Bristol. Bristol have the the foresight to employ Nikki as a full-time enterprise architect, which they should be congratulated on. Nikki has written an excellent report of the workshop on her blog. (Her blog is worth subscribing to, by the way).
What struck me about the day was the variety of architectural patterns that people were pursuing under the general heading of SOA. This echoed our own situation at Edinburgh, where we have got some traction for the basic ideas of service-orientation and now have to articulate our vision for the next steps.
Cardiff University, for example, are basically implementing a real-time data warehouse, with web services making the data available to downstream systems. The data warehouse includes data from the student record, staff database, and finance systems, brought into a common model. It serves as the source of data both for their business intelligence (BI) initiative and for the aforementioned web services.
At Edinburgh, we have a similar but more ad-hoc system for distributing student record data, and we have our own BI intiative which is looking at data warehousing. I had not realised that it was possible to combine the two, as I thought data warehouses were updated nightly or even weekly. I learnt from Cardiff that the technology has now advanced so that the data warehouses can be kept up to date, lagging only a few minutes from the "golden copy" sources.
By contrast, Imperial College London are building a classic SOA architecture, in which multiple business services are orchestrated by the SOA middleware. This is the approach that is most clearly supported by our current SOA technology, Oracle SOA Suite. It is no coincidence that Imperial also use this technology. By contrast, Cardiff started down this route almost a decade ago but have retreated from it, partly because "they don't have services to orchestrate".
A third approach is event-driven SOA, which was exemplified by the University of Lincoln. In this approach, each data update is announced to a range of services, which can take any action they need to. This combines the data-driven aspect of Cardiff's approach with the service-based aspect. It gives a more loosely coupled interation that the centrally orchestrated approach of classis SOA. A central aspect of this approach is an Enterprise Service Bus, which is the mechanism for announcing each event to the relevant services.
At Edinburgh, our discussions often assume we are taking the event-driven approach to SOA. However, we haven't really implemented this yet. We don't have a range of underlying services. Our chosen SOA software is better at supporting the orchestration model, and our IT mindset hasn't really caught up with the idea of calling services rather than moving data. In many cases, we are integrating third-party systems and we really do need to transfer data between systems. So my priority is to clarify these different approaches, explain them to our stakeholders, and decide when we want to use each one.
What struck me about the day was the variety of architectural patterns that people were pursuing under the general heading of SOA. This echoed our own situation at Edinburgh, where we have got some traction for the basic ideas of service-orientation and now have to articulate our vision for the next steps.
Cardiff University, for example, are basically implementing a real-time data warehouse, with web services making the data available to downstream systems. The data warehouse includes data from the student record, staff database, and finance systems, brought into a common model. It serves as the source of data both for their business intelligence (BI) initiative and for the aforementioned web services.
At Edinburgh, we have a similar but more ad-hoc system for distributing student record data, and we have our own BI intiative which is looking at data warehousing. I had not realised that it was possible to combine the two, as I thought data warehouses were updated nightly or even weekly. I learnt from Cardiff that the technology has now advanced so that the data warehouses can be kept up to date, lagging only a few minutes from the "golden copy" sources.
By contrast, Imperial College London are building a classic SOA architecture, in which multiple business services are orchestrated by the SOA middleware. This is the approach that is most clearly supported by our current SOA technology, Oracle SOA Suite. It is no coincidence that Imperial also use this technology. By contrast, Cardiff started down this route almost a decade ago but have retreated from it, partly because "they don't have services to orchestrate".
A third approach is event-driven SOA, which was exemplified by the University of Lincoln. In this approach, each data update is announced to a range of services, which can take any action they need to. This combines the data-driven aspect of Cardiff's approach with the service-based aspect. It gives a more loosely coupled interation that the centrally orchestrated approach of classis SOA. A central aspect of this approach is an Enterprise Service Bus, which is the mechanism for announcing each event to the relevant services.
At Edinburgh, our discussions often assume we are taking the event-driven approach to SOA. However, we haven't really implemented this yet. We don't have a range of underlying services. Our chosen SOA software is better at supporting the orchestration model, and our IT mindset hasn't really caught up with the idea of calling services rather than moving data. In many cases, we are integrating third-party systems and we really do need to transfer data between systems. So my priority is to clarify these different approaches, explain them to our stakeholders, and decide when we want to use each one.
Comments