Online multi-resource allocation for deadline sensitive jobs with partial values in the cloud

In many applications including interactive services and big data analytics, a timely result with a good match is often more valuable than a perfect yet delayed result. This fact can be utilized to improve the total utility gain of a cloud computing platform by allowing partial execution of jobs. A fundamental challenge, however, is that in many real environments, scheduling decisions have to be made online without knowledge about future jobs, which makes it difficult to choose between more valuable jobs with large deadlines and less valuable jobs that are more emergent. Moreover, jobs are often heterogeneous in their utilities, deadlines, and demands for different types of resources. In this paper, we study the problem of online scheduling for deadline-sensitive jobs with concave utility functions that can deliver partial results. We develop efficient online multi-resource allocation algorithms that achieve low competitive ratios for both continuous and discrete job models.