Font Size: a A A

The Research And Design Of A Opportunistic High-throughput Computing Model

Posted on:2006-06-17Degree:MasterType:Thesis
Country:ChinaCandidate:B C ZhangFull Text:PDF
GTID:2168360155465392Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
With the desktop computers becoming more and more powerful, and becoming more and more popular, the unused computer resources are increasing. Especially the distributed computing has got more popularity, many typical environments are that traditional cluster focus on runing program that need plenty of computations all day,and when larger calculation running on a central Linux Cluster, meanwhile there are many Windows PCs idling. So with the distributed computing technology developing, how to utilize idle computing resources have been focused more and more attention. Now many distributed systems have been proposed to utilize these idle computer resources, for example SETI@HOME project project and COW architecture, these projects all running distributed computing when desktop PC is idle. In general these systems characterized as opportunistic high-throughput systems, high-throughput systems mean that they can handling many independent jobs. opportunistic mean that try it best to utilizing as more as possible idle computing resources when no one use PCs . So we combining current several emerging technologies to design a opportunistic high-throughput computing model. Our aim is that idle Windows PC can help core Linux cluster complete high-throughput computing.The method is that applications that initially startup on Linux cluster can transparent migrate to Windows PC and running.First,applications that running on Linux cluster would can transparently migrate to client. We compared implement methods of process migration facility and choose like-FEHAL SSI facility. Second, applications would can execution in non-dedicated clients. Commonly the models that application execution on client can be fall into three main categories: â‘ native execution model. â‘¡sandboxing model. â‘¢specially client program model. In this article, we compared these models, and analyzed two well-know high-throughput systems that Sun Grid Engine and Condor, finally in our model we choose sandboxing model.Finally, through whole model work together those two technology , Windows PC would can help Linux while they are idle. In this article,we describe main architecture of the model, the model consist of three components: master execution node, client, configuration server. Like-FEHAL cluster runnig on master execution nodes,client is Windows PC that deploy with client software, configuration server mainly provide client authorize ,setup and update services.We also discussed problem of network setup on client. We emphasized to described sandboxing environment implement on client . To construct a sandboxing environment,. generally there are three general solutions: â‘ Emulate a complete PC hardware in software and boot a guest operating system on top of it.â‘¡run a specially prepared user mode guest operating system.â‘¢emulate the guest system at the API level. Compared these solutions, we adopt run a specially prepared user mode guest operating system to realize a sandboxing environment. Then,we describe implement of sandboxing low-level and service module. Finally discuss installation and configuration on client.
Keywords/Search Tags:opportunistic high-throughput computing, sandboxing environment, transparent process migration facility
PDF Full Text Request
Related items