Evaluating resource allocation algorithms in clouds

Abstract

Nowadays, cloud services are gaining more and more popular in the world of computing, with the potential of completely replacing on-premises computing resources. The main characteristic of cloud computing services is the ability to share hardware resources between different users, to which amortize the cost of running computations. For the cloud service providers, one of their main focuses is how to optimally minimize the actual physical hardware costs (the machines) that should be spun up, while still be able to fulfill all the computation needs of users. In this project, a job scheduling algorithm is implemented and tested on a public cloud provider’s trace to see the efficiency of it in reducing the hardware costs, in a real-time-simulated environment. A baseline simulation with no algorithm is also implemented and the results show that the algorithm is able to reduce the hardware costs by about 5%.