Output details
11 - Computer Science and Informatics
University of Leeds
An Approach for Characterizing Workloads in Google Cloud to Derive Realistic Resource Utilization Models
<02>Cloud inefficiency is largely attributed to poor understanding of workload and user characteristics, but rigorous analysis and modelling have been lacking. We present a pioneering (IEEE-SOSE'2013 best-paper award) academic effort in analysing very large real-world Cloud data. We combined our expertise in data science (TSB/Rolls-Royce STRAPP, £1.37M) to analyse Google Cloud's month-long operational tracelogs, with our modelling/simulation (ESRC MoSeS/GENeSIS, £2.4M) and middleware (CROWN/iVIC) expertise to develop realistic models for Cloud datacenters. Immediate impact on scientific research into Clouds includes our award-winning work [Xu4] and on-going collaboration with technology leaders: Google (John Wilkes) and Open Science Grid (Prof Miron Livny, HTCondor).