Managing dynamic enterprise and urgent workloads on clouds using layered queuing and historical performance models

David A Bacigalupo,Jano Van Hemert,Xiaoyu Chen,Asif Usmani,Adam P Chester,Ligang He,Donna N Dillenberger,Gary B Wills,Lester Gilbert,Stephen A Jarvis

doi:10.1016/j.simpat.2011.01.007

Abstract

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. We experimentally investigate an historical and a layered queuing performance model and show how they can provide a good level of support for a dynamic-urgent cloud environment. Using this we define, implement and experimentally investigate the effectiveness of a prediction-based cloud workload and resource management algorithm. Based on these experimental analyses we: (i) comparatively evaluate the layered queuing and historical techniques; (ii) evaluate the effectiveness of the management algorithm in different operating scenarios; and (iii) provide guidance on using prediction-based workload and resource management.

Highlights

Being able to make ‘what-if’ response time predictions for potential allocations of enterprise workload to resources, can enhance automatic enterprise ‘workload to resources’ allocation systems [1,2]
The layered queuing method [9] is of particular interest and will be examined further in this paper as it explicitly models the tiers of servers found in this class of application, and it has been applied to a range of distributed systems (e.g. [10]) including the benchmark used in this paper [11]
We have found in previous experiments that there are three important operating scenario variables to cover when investigating the effectiveness of a prediction-based management system: 1) the extent to which the system is loaded; 2) the scenario heterogeneity; and 3) whether the performance predictions are tuned to compensate for predictive inaccuracy [4,20]

Summary

Introduction

Being able to make ‘what-if’ response time predictions for potential allocations of enterprise workload to resources, can enhance automatic enterprise ‘workload to resources’ allocation systems [1,2] These predictions can help provide more cost-effective response time-based Service Level Agreement (SLA) management [22]. Response time predictions can Preprint submitted to Simulation Modelling Practice and Theory be very useful when enterprise systems are run on clouds or similar resourcesharing infrastructures These clouds are often very large, increasing the number of possible workload-resource allocations and providing more opportunities for the use of predictions. Examples of the first approach include the use of both coarse [3] and fine [1] grained historical performance data. A layered queuing model can be solved either by simulation or via an approximate analytical solver

Methods

Results

Conclusion