PERSISTENCE PRIORITY QUEUE BASED JOB SCHEDULING

Abstract

This thesis considers the problem of performance degradation of servers caused due the sudden increase in the requests resulting in request rejection and performance degradation. This overload condition can occur at any time without any predication. During this critical condition of overload, performance criteria such as response time and throughput get hampered which in turn affects user satisfaction level and can lead to imploded revenues. Some users are important than others and hence they require special care under these overload conditions. We can distinguish these important customers by assigning priorities to them. Selecting higher priority user can be done using priority queue but simple priority queue implementation are non- persistent in nature and hence does not resist failures and it is difficult to recover or recovery may take a long time. Hence, we need a solution so that these requests become persistent in nature and do not get lost on failures and get easily recovered. We have proposed the database implementation of queue such that it selects the highest unprocessed job for processing. Since requests are stored in database and this database can be made highly available resulting in approximate persistent system. To overcome starvation problem and to make scheduling algorithm more responsive while handling user requests with fixed priority, we have proposed a dynamic priority scheduling algorithm where the priority of the request is determined at execution time and it is weighted combination of priority scheduling and shortest job first with aging concept. Our algorithm is also motivated towards better profit for service provider while maintaining good response time and throughput. Our simulation results show that our strategy can provide good profit to service providers. Our results also show that higher priority jobs are handled with special care and low priority jobs do not get starved.