Research On Available Technologies For Job Scheduling In The Grid

Driven by increasingly complex problems and propelled byincreasingly powerful technology,today's science is as much based oncomputation, data analysis, and collaboration as on the efforts ofindividual experimentalists and theorists. But even as computer power,data storage, and communication continue to improve exponentially,computational resources are failing to keep up with what scientistsdemand of them.scientists try their best to find one kind of computingmodel which is cheapness and powerful, at last they find the answer-Gridcomputing.Grid computing, most simply stated, is distributed computingtaken to the next evolutionary level. The goal is to create the illusionof a simple yet large and powerful self managing virtual computer outof a large collection of connected heterogeneous systems sharingvarious combinations of resources.With the emergence of grid computing,because of resources in gridwhich belong to different organization or individual isheterogeneous ,geography distributed,the schedulers can not manage theresources directly,new challenges appear in grid job scheduling.new jobscheduling models and scheduling algorithms are in need to schedule thejobs in grid. The job schedule algorithms must have the ability to absorbthe dynamic property of resources in grid ,such as the host add , removeor variety demand of resource provider .at present ,most of job schedulealgorithms use prediction information on the underlying resources andjobs,load balancing is not considered in these algorithms.ingrid ,resource can easily under useless situation which is caused byunpredicted reason because resources are managed by the resourceproviders themselves.at this time ,the jobs running under the downtimeresource must migrate to other resource .In order to overcome this drawback,in this paper we proposehierarchically dynamic load balancing system which can improve utilizeof computing resources and throughput of task.The emphasis of this paper is:studying grid technology and gridservice,designing and realizing the hierarchically dynamic loadbalancing system in order to absorb the heterogeneous and dynamicproperty of grid resource.we have also proposed SMM dynamic loadbalancing algorithms and migrate jobs under the downtime resource.Details as follow:(1) study Grid Computing and Grid service Theory;summarizing therelated concepted;study the notions of OGSA and Grid service;In-depthstudy grid job management.(2)design and implement hierarchically dynamic load balancingsystem in GridSim simulation,which is consist of:Index service:get information from grid resource ,includingworkload information.it is very important that we can get the gridresource information fast and accurately.so it can provide the cogentreference for matching the resource and workload information for theload balancing.Resource matcher:get the available resource information from indexservice base on the job description,submit all these information to thescheduler.Scheduler:use allocating policy to schedule the jobs base on theinformation of jobs and resources.Load balancer:computing gain and cost of the load balancing,loadbalancer would invoked when gain is bigger than cost.(3) design and propose segmented Min-min hierarchically dynamicload balancing algorithm.Hierarchically dynamic load balancing algorithm:at first,we usethe segmented Min-min algorithm to assign the jobs,then the globalscheduler periodically (interval is d(h)) computing the gain and costof load balancing to decide whether do the load balance or not.in thisinterval,the regional Scheduler also do the computationperiodically(interval is d(l))to decide Whether do the load balance ornot,this is similar to the global scheduler.In this algorithm we will introduce load informationpredicting,task scheduling algorithm,the gain and cost of loadbalancing.Load information predicting :get the load information from indexservice to predict future load information.Task scheduling algorithm:Min-min is a simple algorithm,fast,andable to deliver good performance.the drawback of Min-min is that itassigns the small job first.Thus,the smaller jobs would execute firstand then a few larger jobs execute while several machines site idle,resulting in poor matching utilization and load unbalancing.in thispaper we propose the segmented Min-min algorithm which comes fromMin-min.the job list is partitioned into segments with equal size.foreach segment,Min-min is applied to assign jobs to machines.Gain and cost of load balancing:in this paper we use a heuristicalgorithm to computing gain of load balancing ,the cost of loadbalancing including state save,state recovery and communicationoverhead.(4)design and propose a load-based job migration policyPlanning based resource management systems are widely used in thearea of Grid computing.advance reservations are important allocationstrategy,widely used,e.g.,in Grid toolkits such as Globus,as theyprovide simple means for planning of resources and in particularco-allocations of different resources. we use advance reservation tobook the resource in order to complete job in time.how to select jobswhich have been submitted to the broken resource is a big problem.whilecurrent research mainly focused on recovery mechanisms for those jobsthat are already active.in this paper we will migrate jobs which areaffected but not yet active .people usually estimate the actual downtimeof resource to decide which jobs must be migrated,underestimate theresource down time will result in an increased amount of jobtermination.so we propose a load-base job migration strategy,whichcalculate the remapping interval to decide which jobs must be migrated.