Research On Task Management Mechanisms For Multiple-Discipline Virtual Experiment

With the development of information technology, the virtual experimental techniquesas a research tool have become more and more significant in various disciplines. But thereis no uniform standard for traditional single-discipline virtual experiment systems, whichleads to the various implementations and poor compatibility among different systems.Based on the need of Multiple Disciplines Virtual Experiment Platform, amanagement system that is considered to be a core subsystem has been designed andimplemented. The system solves the problem with supporting large scale concurrent taskrequests. And it provides a set of secure and unified application interfaces to the upperlayer and can communicate with the underlying cluster computing nodes. Besides, thesystem also completes a range of services for the other subsystems to access. All of theseguarantee the efficiency, stability, reliability and scalability of the platform. Analysing theactual needs of the subsystem, three optimization strategies are proposed. The subsystemis mainly based on Twisted, which is a free web service framework realized in Python, andsome improvements to the framework are adopted to support large-scale concurrency. Inorder to increase the reliability and availability of the system, a task-migration mechanismbased Modelica is proposed to tackle some service failures. Meanwhile, a NAT penetrationmodule of TCP based STUNT protocol is added to reduce the amount of data through themaster node. Besides, the system provides a package of functional modules, such asdata-compression, resuming broken transfer and some other service interfaces. Theadvantage of the resuming broken transfer is to solve the problem that the temporarilynetwork disconnecting or instability between the client and the server leads toexperimenting interruptions. If the data-compression functionality switch turns on, theamount of data transferred on the network is reduced greatly and the transfer latency isalso decreased.The system is implemented based on fully analyzing the system requirements anddigging out the deep characteristics. Experimental results show that the system can be ableto provide reliable service. Compared with the original Twisted framework, the improvedmulti-process model can increase the number of concurrent connections by more than50%in a given response time. And with the help of the NAT traversal of TCP module, thedata-flow through the master node can be reduced by30%or more.