Online Algorithms for the Interval Scheduling Problem in the Cloud: Affinity Pair Threshold Based Approaches

In the interval scheduling problem, jobs have known start and end times (referred to as job intervals) and must be assigned to processing nodes for their whole duration. Although the problem originally stems from the resource allocation demands of resident processes in operating systems, it found a renewed interest in the Cloud context, both in IaaS and SaaS, since reservations for virtual machines and services often have known activation intervals. A common objective of interval scheduling is to minimize busy time of machines which relates (among others) to minimizing the number of machines participating in the computation. As a consequence, bin packing techniques have been applied in the past. In this paper we tackle the online version of the problem, whereby future job arrivals are unknown. We propose novel algorithms that work as a pre-processing step to any bin packing scheme by offering recommendations that are enforced in all packing decisions. Job overlaps are used to characterize pairwise job affinity and subsequently provide threshold based job allocation recommendations. Thresholds are calculated using lower bound theoretical analysis upon two extreme workloads (sparse and dense). Experimental evaluation using real world workloads illustrates the merits of our approach against state-of-the-art algorithms.

Automated summary

This paper discusses the online version of the interval scheduling problem and proposes a novel algorithm for preprocessing bin packing schemes and providing job allocation recommendations based on job overlaps. Experimental results show the advantages of this approach compared to existing algorithms.