Scheduling with advanced reservations

Abstract

Some computational grid applications have very large resource r equirements and need simultaneous access to resources from more than one parallel computer. Current scheduling systems do not provide mechanisms to gain such simultaneous access without the help of human administrators of the computer systems. In this work, we propose and evaluate several algorithms for supporting advanced reservation of resources in supercomputing scheduling systems. These advanced reservations allow users to request resources from scheduling systems at speci c times. We nd that the wait times of applications submitted to the queue increases when reservations are s u p p orted and the increase depends on how reservations are supported. Further, we nd that the best performance is achieved when we assume that applications can be terminated and restarted, back lling is performed, and relatively accurate run-time predictions are u s e d.

Deadline-sensitive workflow orchestration without explicit resource control

“…Thus advanced reservations yield resource slots that have guaranteed start and end times and probability of procurement very close to one. However, advanced reservations mechanisms have shown to have a negative impact on utilization and hence resource providers often allow only limited use of this feature [27,28]. In addition, TeraGrid sites require requests to be made at least 24 to 48 hours in advance, which is not practical for applications with dynamic loads.…”

“…Resource mechanisms and protocols are available today to coordinate grid resources and ensure quality of service (QoS) [5,6,27]. There are tools for workflow planning using performance models [3,10,18] and execution systems or workflow engines for managing runtime environment of workflows [4,15].…”

An RMS Architecture for Efficiently Supporting Complex-Moldable Applications

“…Advance reservations are used as they offer a solution to the co-allocation problem [20], however, they also fragment resources [23], thus leading to inefficient resource utilisation. Redundant requests are multiple jobs targeted at individual clusters: when one of these jobs starts, the others are cancelled.…”

Overdimensioning for Consistent Performance in Grids

“…Previous work focuses either on providing predictable performance using advance reservations or prediction methods [3], [4], [5], or meeting the performance requirements specified by Service Level Agreements (SLA) [6], [7]. In contrast, in this study we present a detailed investigation of overdimensioning as a solution for achieving consistent performance in grids.…”