On Advantages of Grid Computing for Parallel Job Scheduling

Communication-Aware Job Placement Policies for the KOALA Grid Scheduler

“…However, co-allocation is not without cost; although it is expected to reduce the average job response time [12], it has the potential drawback of increased execution times of applications due to low network bandwidth and high latency over wide-area networks. Hence, the performance of co-allocation may be severely affected by the slow inter-cluster connections.…”

“…If the placement succeeds, this means the total processors requested of the job is satisfied; hence the job is placed successfully. Otherwise, if the processor request is more than what is available in the cluster considered, it places a component with a size equal to the number of idle processors of that cluster, and subsequently, updates the number of processors requested for the remaining number of requested processors (line [10][11][12][13][14]. After traversing the complete ordered cluster sequence, if the processor request cannot be satisfied, the job is put into the placement queue of KOALA (lines [16][17][18].…”

A Dynamic Co-allocation Service in Multicluster Systems

“…There, we considered both synthetics workloads, and workloads derived from the logs of the DAS and from application runtimes on the DAS. In [11,12], co-allocation (called multi-site computing there) is studied with simulations, with as performance metric the average weighted response time. One of the most important findings is that when the slowdown of jobs due to the wide-area communication is less than or equal to 1:25, it pays to use co-allocation.…”

Scheduling parallel batch jobs in grids with evolutionary metaheuristics

“…Many studies propose either a distributed management system (Ernemann and Yahyapour 2003) or centralized scheduling (Ernemann et al 2002). There are also combinations of distributed and centralized management (Vazquez-Poletti et al 2007) that can be characterized as a type of hierarchical multilayer resource management system (Schwiegelshohn et al 2008).…”