A Queuing Model to Achieve Proper Elasticity for Cloud Cluster Jobs

Abstract

Achieving proper elasticity for cloud jobs is a challenging research problem and far from being solved. In this extended and revised paper, we investigate how to achieve proper elasticity for highly parallelized jobs which run on cloud clusters. In particular, we present an analytical model based on finite queuing systems that can be used to determine at any given instance of time and under current workload conditions the minimal number of cloud resources needed to satisfy the SLO requirements such as response time and request loss probability. We give numerical examples to demonstrate the applicability and usefulness of our model. In particular, we study the performance of a cloud-based clustering system in terms of queue utilization, request loss, and response service time. We also give an example of fluctuating workload to show how our model can be used to achieve proper elasticity. Discrete Event Simulation has been used to verify the correctness of our proposed model.

A contribution to scheduling jobs submitted by finite-sources in computational clusters

“…The common queue scheme can be used in MaaS (Message Queueing as a Service) in the cloud computing paradigm. A possible example of the computeintensive services of unknown service times is AWS cloud lambda service [4,5].…”