Scalability Analysis of an Enterprise Service Bus

Abstract

This master's thesis investigates the scalability of an Enterprise Service Bus (ESB) made by Kantega for SpareBank 1. The ESB is written in Java and its performance is closely related to the workings of the Java Virtual Machine (JVM). The scalability analysis is done by following the \textit{Analysis and Scalability Method} which is adapted for this thesis by combining two existing scalability analysis methods. Unlike most scalability analyses the aim was to examine the effects on the performance also when scaling down by giving the ESB limited access to the hardware resources of the server. The system was tested on four different system sizes: double server, one server, half server and quarter server. The main goals were to analyse the scalability, determine if scaling down has predictable effects on the performance, and consider if the \textit{Analysis and Scalability Method} was fitting for analysing the scalability of the system.Scaling the system showed linear results when scaling up to multiple servers, but when scaling down the results were non-linear as it proved to give better performance when the system were limited to fewer cores. The system showed predictable behaviour when it was scaled down to a single processor, but scaling a system down to a single core within a multi core processor is not recommended as the distribution of shared processor level caches can not be predicted. The \textit{Analysis and Scalability Method} proved to be a good approach for analysing scalability, although it may be too labour intensive for smaller projects.