Applying the heterogeneity level metric in a distributed platform

Heterogeneity Level (HL) metric has been developed by our research-group to help scheduling algorithms to adapt themselves to the existent heterogeneity in the platforms. This paper presents our results considering the HL’s behaviour in a real adaptive scheduling. HL metric quantifies qualitative aspects from heterogeneity in order to provide efficient performances and lower cost to the execution in both heterogeneous and homogeneous platforms. HL use is investigated under different perspectives:CPU, memory, network and considering benchmarks results. A simple but effective adaptive scheduling using HL is proposed and its results point out to performance-gains around 53% when a non-adaptive scheduling algorithm is used. Our case studies show that the HL was efficient, flexible and easily used for scheduling policies.