Study Of Network Communication Optimizationinter Inter VMs

Hardware enhancement brought the development of virtualization technology,which makes cloud computing better availability.Virtualization technology is applied to organize and manage the computing resource in cloud computing environment,which introduce virtual machine as the basic resource encapsulation.And virtualization technology is also one of the basic technologies of IaaS(Infrastructure as a service)-enabled in cloud computing environment.With the universalness of cloud computing,business expansions promote industrial circles?academic circles to research how to provide better cloud computing services in turn.Better cloud computing services means faster computation speed?larger storage and higher network communication performance,which are basics and cores of cloud computing services.Virtual applications in cloud computing spread over HPC?large-scale distributed computing?Web transactions processing,network communication is one of the most important workload of them.It's very necessary to improve network communication performance for network-intensive application in cloud computing.Co-located VMs is defined as VMs which are located on the same physical machine.Study of network communication optimizationiter via co-located VMs,has also been a hot pot of industry and academia.It's very significant for the particle value of a network communication optimizer inter VMs to not only improve network performance,but also support live migration and flexibly deployment,further more assure application transparency.So it's very important to study a network communication optimizer inter VMs with peculiarities as follows:(1)Improving the performance of network communication via co-located VMs by supporting inter-VMs communication based on co-location awareness:Predicting whether both of communication parties are co-located.If yes,the pattern of network communication shoud be switched to local pattern based on share memory connection,to shorten the path of network communication via co-located VMs? reduce network overhead,thus to cut extra loss brought by resources vitualization;If not,the pattern of network communication shoud beswitched to remote pattern based on TCP/IP.(2)Supporting Seamless Agility: Seamless Agility is one of important characters of virualization technnogy.Seamless Agility means migrating living VMs from one physical machine to another physical machine without breaking in services,which can be used to support system workload balancing ? fault tolerance-recovery? energy reduction and cloud computing cluster management improvement.The desination of good optimizing mechanisms of network communication via co-located VMs shoud not brek this character,which need to support transparentswitch of local patern based on sharing memory connection and remote pattern based on TCP/IP.(3)Assurance of “Application-OS Kernel-VMM” multi-level transparency: multi-leveltransparency means no need to add new reductant program interfaces?no need to modify legacy applications?no need to recompile program and load different libraries.This character can improve the mechanism's usability and feasibility significantly.(4)Supporting TCP and UDP semantic: TCP and UDP are most widely used in network applications.A good network communication optimizationinter co-located VMs shoud support them both.Focusing problems above,this article aims at designing and implementing a network communication optimizer inter VMs base on sharing memory connection-XenVMC,which replaces TCP/IP by share memory connections as network communication channel inter co-located VMs.This optimizer improve network communication performance?support TCP and UDP?support seamless agility?assure “Application-OS Kernel-VMM” multi-level transparency.In a word,this thesis has contributions bellow:(1)In consideration of Xen and Linux are most widely used in academia and industry,this thesis has designed and implemented a a network performance optimizing mechanism via co-located VMs base on sharing memory connection based on Xen and Linux,which is implemented on system call level?bypass network communication of co-located VMs,achive high performace,assure “Application-OS-VMM” multi-level transparency,support TCP and UDP semantics by intercept sockets' type.We have tested it with other mechanisms on the same platform to verificate our work.The experiment testified,our optimizationinter's UDP transaction processing performance is 10.9x compared with netfront/netback's?1.9x compared with XenLoop's at most;its TCP transction processing performance is 11.9x compared with netfront/netback's?1.8x compared with Xen Loop's at most;its UDP data stream processing performance is 9.76 x compared with netfront/netbakc's?3.2x compared with Xen Loop's;its data stream processing performance is 3.27 x compared with netfront/netback's?1.6x copared with XenLoop at most.(2)This thesis designed an algorithm of co-located VMs set dynamic maintenance based on event-driven,which is developed by sdudy the change law of co-located VMs set in cloud computing cluster.Compared with algorithm of post co-located VMs set dynamic maintenance based on round robin scheme,it cost lower and responde faster.(3)We have implemented the algorithm above in the communication optimizationinter,which makes the cost of co-located VMs set maintenance lower,and supporting seamless agility assurance.The experiment testified this mechanism support seamless agility,the legacy data brought by VM immigration and VM emigration is handled correctly.