Section: New Results

Cloud Computing

Even though a new era of Cloud Computing has emerged, the characteristics of Cloud load in data centers is not perfectly clear. In [20] , we characterized the job/task load and host load in a real-world production data center at Google Inc. by using a detailed trace of over 25 million tasks across over 12,500 hosts. We found that the Google data center exhibits finer resource allocation with respect to CPU and memory than that of Grid/HPC systems and Google jobs are always submitted with much higher frequency and they are much shorter than Grid jobs, leading to higher variance and noise. Moreover, as far as prediction is concerned, we designed in [21] a Bayes model to predict the mean load over a long-term time interval, as well as the mean load in consecutive future time intervals. Real-world experiments showed that our Bayes method achieved high accuracy with a mean squared error of 0.0014 and that it improves the load prediction accuracy by 5.6-50% compared to other state-of-the-art methods based on moving averages, auto-regression, and/or noise filters.

In a similar vein, the exploitation of Best Effort Distributed Computing Infrastructures (BE-DCIs) allows operators to maximize the utilization of the infrastructures, and users to access the unused resources at relatively low cost. Profiling the execution of Bag-of-Tasks (BoT) applications on several kinds of BE-DCIs demonstrates that their task completion rate drops near the end of the execution. In [33] , we presented the SpeQuloS service which enhances the QoS of BoT applications executed on BE-DCIs by reducing the execution time, improving its stability, and reporting to users a predicted completion time. We presented the design and development of the framework and several strategies to decide when and how Cloud resources should be provisioned; moreover, performance evaluation using simulations showed that SpeQuloS fulfill its objectives in speeding up the execution of BoTs, in the best cases by a factor greater than 2, while offloading less than 2.5% of the workload to the Cloud. These topics were also further explored in the book chapter [30] .