We all know the theory of distributed computing: by dividing tasks among several computers rather than having all processes originating from one central computer, we can increase overall throughput. The problem is that in reality the actual implementation of such a scheme is often quite complex.

Technologies like EJB were supposed to make it easier, but they proved extremely intrusive to both the design and development process. Fortunately, the emergence of JVM-level clustering technologies like Terracotta into the mainstream now provides a viable alternative.

Recently Shine Technologies distributed one of its applications using Terracotta and achieved significant performance gains. In the past, our application performance was generally limited by the server it was running on. With Terracotta this no longer appears to be the case. During our performance testing, when one server was at capacity, we just added another and the overall application throughput increased markedly - to the point that our main limiting factor was the database.

Some Background

Shine Technologies is a provider of IT consultancy and development services to a number of companies in the Australian energy industry. Following the adoption of Full Retail Contestability (FRC) in this industry, Shine has developed numerous products for these companies.

In a nutshell, these products help facilitate the financial interactions between distributors and retailers regarding network usage. These interactions are very high volume, with large retailers having to process millions of transactions per month. Consequently, the ability of the application to scale becomes business-critical - particularly when all of the transactions are received by the retailer on the same day.

A recent addition to this suite of products has been an application to be used by a service provider to electricity retailers. This application - known as the Market Reconciliation System (MRS) - receives large volumes of electricity usage data each week, and provides a reporting and reconciliation mechanism to retailers, highlighting any discrepancies between what they expect to be charged for usage and what they are actually being charged.

Given the high-volume nature of this product, scalability was a big driver in the application architecture. After an initial proof-of-concept involving various distributed computing frameworks, Terracotta was deemed to be the framework most likely to give MRS the grunt it needed.