Resource management techniques for handling uncertainties in user estimated job execution times

The popularity of clouds is growing rapidly. Research on cloud computing has stared considering Service Level Agreement (SLA) characterized by an earliest start time, runtime and a deadline for completion with a request for job execution submitted to the cloud. Estimates of request runtimes provided by users are often error prone. Both overestimation and underestimation of request runtimes are detrimental for system performance. Satisfying SLAS often require Advance Reservation (AR) of resources. This research presents a Soft Advance Reservation (SAR) approach for handling errors associated with estimates of request runtime provided by users. This approach relaxes the strict requirement that all Advance Reservation (AR) requests associated with their respective SLAs must meet their deadlines. Two algorithms based on the SAR approach that adjust the user provided estimates for improving system performance are discussed. A simulation-based performance evaluation is used to demonstrate the effectiveness of these algorithms.