Evaluating container scheduling in clouds

Abstract

There has been significant progress in cloud scheduler research. Cost-aware schedulers are cloud schedulers that aim to reduce the cloud costs by packing as many tasks as possible onto some number of instances. Stratus is a cost-aware scheduler developed by Andrew et al. to pack tasks efficiently to reduce cloud costs. This report compares Stratus against a baseline algorithm where one task is mapped to one instance using a round-robin scheme to select instance types. An experimental scheduler is built to feed a historical Microsoft Azure Compute virtual machine (VM) request dataset into both algorithms to compare their costs. Metrics like number of instances, instance core utilisation, memory utilisation, and average maximum task runtime of the instance pool were collected. After conducting the experiments, it is found that Stratus outperforms the baseline algorithm in reducing cloud costs.