Preemptive Job Scheduling with Priorities and
Starvation Avoidance CUM Throughput Increasing
Tool in Clusters
International Journal on Communication, 1 (3):
4 (December 2010)Abstract
This paper proposes a new scheduler to schedule
parallel jobs on Clusters that may be part of a Computational
Grid. This proposed policy proposes 3 Job Queues. In each
Cluster, the first Queue has some jobs which are from
Computational Grids. It means, it may be either bigger job or
part of the bigger job from the Computational Grids. The
second Queue has jobs which has low required execution time.
The third Queue has jobs which has high required execution
time. In first and second Queues, there is no chance of
starvation. But in third Queue, there is a chance of starvation.
So this proposed policy applied AGING technique to preempt
the jobs which has low priority. And the first Queue is fully
dedicated to execute a part of bigger jobs (Meta-Jobs) only. So
here we maintain three job Queues which are effectively
separate jobs according to their required execution time for
local Jobs, bigger job/part of bigger job(Meta-Job) from
Computational Grids. Here we preempt jobs by applying
AGING Technique. Initially 20% of total available resources
(processors) will be allocated to first Queue only. And 40% of
total available resources (processors) will be allocated to
second and third queues respectively. Whenever the third
queue is Empty then the proposed scheduler simply selects the
job which has least required execution time and executes it
immediately. In this way, the proposed scheduler increases the
throughput in clusters.