Year: 2016
Type: dissertation
Abstract: Nowadays, clustered environments are commonly used in enterprise-levelapplications to achieve faster response time and higher throughput thansingle machine environments. However, this shift from a monolithic architecture to a distributed one has augmented the complexity of these applications, considerably complicating all activities related to the performance optimisation of such clustered systems. Therefore, automatic techniques are needed to facilitate these performance-related activities, which otherwise would be highly error-prone and time-consuming. This thesis contributes to the area of performance optimisation of clustered systems in Java (a predominant technology at enterprise-level), especially aiming for large-scale environments. This thesis proposes two techniques to solve the problems of efficiently identifying workload-dependent performance issues and efficiently avoiding the performance impacts of major garbage collection, two problems that a typical clustered Java system would likely suffer in large-scale environments. In particular, this thesis introduces an adaptive framework to automate the usage of performance diagnosis tools in the performance testing of clustered systems. The aim is to ease the identification of performance issues by decreasing the effort and expertise needed to effectively use such tools. Additionally, an adaptive GC-aware load balancing strategy is introduced, which leverages on major garbage collection forecasts to decide on the best way to balance the workload across the available nodes. The aim is to improve the performance of a clustered system by avoiding the impacts in the cluster's performance due to the major garbage collection occurring at the individual nodes. Experimental results of applying these techniques to a set of real-life applications are presented, showing the benefits that the techniques bring to a clustered Java system.
Authors: Portillo Dominguez, Andres Omar
Language: English
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
Funder: Science Foundation Ireland
OpenAlex ID: W2623141317