Latency-Aware Empirical Analysis of the Workloads for Reducing Excess Energy Consumptions at Cloud Datacentres

Abstract

Witnessing the growing importance of energy efficient computing in Information and Communication Technology (ICT), Cloud Computing is certainly being addressed as massive consumers of energy in the recent years. Resource management benefitted from predicting the anticipated near future workloads is one of the suggested strategies for the interests of energy efficiency. But the reliability of this prediction is always being a concern as it is affected by the complexities and constrains imposed by the heterogeneity and dynamism of both the Cloud workloads and the datacentre environments. So, a precise and clear understanding of the properties, characteristics and behaviours of such Cloud entities is of absolute importance. One of the unexplored characteristics of the Cloud Workloads is the latency sensitivity of these workloads, which has a considerable impact on their actual behaviours at the Cloud processing environments. Apart from the communication related latencies such as network latency and dispatching latency, the in-house computational latency of the workloads have not gained sufficient emphasis in the existing works of energy efficient Cloud Computing. To this end, this paper empirically dwells into the effects and impacts of this computational latency up on the behaviours of the Cloud workloads. Extensive analysis has been conducted both at inter and intra-job levels based on the Google trace logs spanning across a period of one month. Further, the constraints imposed by this computational latency on both the end user satisfaction and on the process-level complexities faced by the service providers have been manifested by our experimental analysis, ultimately to drive insightful inferences for reducing the excess and undesirable energy consumptions at the Cloud datacentres.