Font Size: a A A

The Multi-core Multi-threaded Parallel Design And Development Of NS2

Posted on:2017-01-28Degree:MasterType:Thesis
Country:ChinaCandidate:Z R TangFull Text:PDF
GTID:2308330491951690Subject:Computer technology
Abstract/Summary:PDF Full Text Request
With the emergence and rapid development of multi-core processors, it makes sense to parallel the classic serial programs and study on multi-threaded task scheduling problem for the parallel program in order to take advantage of multi-core architecture to improve performance. NS2 is a typical network simulator based on single-threaded design. When carrying out large-scale network simulation, the single-threaded processor core can not meet the requirements of huge computing power. The efficiency of the network simulation becomes very low. You can take advantage of multi-core and multi-thread related technology to improve the efficiency and reduce the time it takes for large scale network simulation.At first, this paper proposed a parallel transformation design scheme of the typical network simulation software NS2 and implement it by using the OpenMP. Extend the event class of object classes in NS2 and add an event pool in scheduler class. Then focus on original serial event scheduling and distribution mechanism of NS2 scheduler, add getting events parallelly module, identifying parallel event module and distributing events parallelly module, in order to achieve parallel processing of events in scheduler. Finally, resolve conflict after parallelization and verify the correctness of scheme. And then,this paper studies on multi-threaded task scheduling problem which occurs when the NS2 is executed in parallel. From the perspective of performance optimization, this paper proposes a multi-threaded load balancing scheduling policy for multi-core platform. Then judge whether a task is suited for multi-threaded parallelization tuning according to the size of each task block. Then deal with the task in two different ways which is suited for multi-threaded parallelization tuning.If the decomposition task block is uniform, use general static scheduling policy. If the task block is uneven, this paper designs a heuristic scheduling algorithm.Finally, based on the parallel transformation design scheme of NS2 proposed in this paper, the simulation experiment and accelerated performance testing is conducted. The results can verify the correctness of the scheme and shows the parallelization and performance tuning are very effective.
Keywords/Search Tags:NS2, multi-core architecture, OpenMP, task scheduling, load balance
PDF Full Text Request
Related items