Escalonamento justo em infraestruturas de nuvem com múltiplas classes de serviço

Cloud computing providers offer multiple service classes to deal with workload heterogeneity. Classes are distinguished by their expected Quality of Service (QoS), which is defined in terms of Service Level Objectives (SLO). A priority-based scheduling policy is commonly used to guarantee that requests submitted to the different service classes achieve the desired QoS. However, the QoS delivered during resource contention periods may be unfair to certain users. In this paper, we present a SLO-driven scheduling policy which takes the SLOs and actual QoS delivered for each request into account when making decisions. We used simulation experiments fed with traces from a production system to compare the SLO-driven policy with a priority-based one. In general, the SLO-driven policy delivered a better service than the priority-based one.