基于超算環(huán)境的面向多租戶的輕量級虛擬HPC集群的設(shè)計與實現(xiàn)
發(fā)布時間:2018-09-01 08:38
【摘要】:為了給用戶提供按需使用的HPC服務(wù)并解決用戶應(yīng)用部署的軟件依賴性問題,在不破壞現(xiàn)有超大規(guī)模高性能集群管理的前提下,基于Fat-Tree網(wǎng)絡(luò)拓撲和虛擬集群模型,通過對部分資源進行云化管理,設(shè)計并實現(xiàn)了一種輕量級的HPC集群交付模式。從而為面向租戶需求的高性能應(yīng)用提供一種云化的HPC租用服務(wù),并解決了HPC應(yīng)用軟件棧的僵化問題,以易用的方式為更多的HPC租戶提供服務(wù)。實驗結(jié)果表明,該方法以近似物理節(jié)點的性能使HPC具備了云的按需使用的特性,并具有較為理想的植入效率。
[Abstract]:In order to provide users with on-demand HPC services and to solve the problem of software dependency of user application deployment, Fat-Tree network topology and virtual cluster model are used to solve the problem without destroying the existing super large scale and high performance cluster management. Through cloud management of some resources, a lightweight HPC cluster delivery mode is designed and implemented. Thus it can provide a cloud HPC rental service for high performance applications to meet the needs of tenants, and solve the rigid problem of HPC application stack, and provide services to more and more HPC tenants in an easy-to-use manner. The experimental results show that the proposed method can approximate the performance of physical nodes and make the HPC have the characteristics of cloud on demand, and it also has an ideal implantation efficiency.
【作者單位】: 國防科技大學(xué)計算機學(xué)院;中南大學(xué)信息科學(xué)與工程學(xué)院;
【分類號】:TP393.09
本文編號:2216659
[Abstract]:In order to provide users with on-demand HPC services and to solve the problem of software dependency of user application deployment, Fat-Tree network topology and virtual cluster model are used to solve the problem without destroying the existing super large scale and high performance cluster management. Through cloud management of some resources, a lightweight HPC cluster delivery mode is designed and implemented. Thus it can provide a cloud HPC rental service for high performance applications to meet the needs of tenants, and solve the rigid problem of HPC application stack, and provide services to more and more HPC tenants in an easy-to-use manner. The experimental results show that the proposed method can approximate the performance of physical nodes and make the HPC have the characteristics of cloud on demand, and it also has an ideal implantation efficiency.
【作者單位】: 國防科技大學(xué)計算機學(xué)院;中南大學(xué)信息科學(xué)與工程學(xué)院;
【分類號】:TP393.09
【相似文獻】
相關(guān)期刊論文 前2條
1 沈軍;滿家巨;聶作先;;高性能集群管理與優(yōu)化[J];計算機與現(xiàn)代化;2007年02期
2 陶華亭;高性能集群環(huán)境中的聯(lián)網(wǎng)效率[J];電腦知識與技術(shù);2004年23期
相關(guān)碩士學(xué)位論文 前1條
1 何子龍;基于回歸分析的多租戶應(yīng)用資源消耗技術(shù)研究[D];廣西大學(xué);2015年
,本文編號:2216659
本文鏈接:http://sikaile.net/guanlilunwen/ydhl/2216659.html
最近更新
教材專著