You are here

Resource management of enterprise cloud systems using layered queuing and historical performance models

TitleResource management of enterprise cloud systems using layered queuing and historical performance models
Publication TypeConference Paper
Year of Publication2010
AuthorsBacigalupo, DA, van Hemert, J, Usmani, A, Dillenberger, DN, Wills, GB, Jarvis, SA
Conference NameIEEE International Symposium on Parallel Distributed Processing

The automatic allocation of enterprise workload to resources can be enhanced by being able to make `what-if' response time predictions, whilst different allocations are being considered. It is important to quantitatively compare the effectiveness of different prediction techniques for use in cloud infrastructures. To help make the comparison of relevance to a wide range of possible cloud environments it is useful to consider the following. 1.) urgent cloud customers such as the emergency services that can demand cloud resources at short notice (e.g. for our FireGrid emergency response software). 2.) dynamic enterprise systems, that must rapidly adapt to frequent changes in workload, system configuration and/or available cloud servers. 3.) The use of the predictions in a coordinated manner by both the cloud infrastructure and cloud customer management systems. 4.) A broad range of criteria for evaluating each technique. However, there have been no previous comparisons meeting these requirements. This paper, meeting the above requirements, quantitatively compares the layered queuing and (\^A¿HYDRA\^A¿) historical techniques - including our initial thoughts on how they could be combined. Supporting results and experiments include the following: i.) defining, investigating and hence providing guidelines on the use of a historical and layered queuing model; ii.) using these guidelines showing that both techniques can make low overhead and typically over 70% accurate predictions, for new server architectures for which only a small number of benchmarks have been run; and iii.) defining and investigating tuning a prediction-based cloud workload and resource management algorithm.

Full Text