Research On I / O Bottleneck Solution Of Cloud Platform Network Based On Load Balancing

With the rise of cloud computing, which is a quite new computing mode, people are seeing a new future brought by it, this is a model that aims at providing scalable virtualized resources as services via the Internet to users, who do not need to know how to manage the underlying infrastructure.The construction and management of the next-generation data center, which is the carrier of the core information of a cloud, is facing a variety of needs and challenges。The new generation data center has brought us benefits like reducing operation costs, improving application compatibility, accelerating application deployment, improving service availability, making resource scheduling dynamic, also reducing energy consumption. However many enterprises are not willing to adopt this computing mode due to security issues as well as the I/O bottleneck. According to a Yankee Group report, many IT leaders think that network latency and bad performance are preventing enterprise to adopt cloud computing. Virtualization has indeed brought enterprise better resource utility and more efficient management, which means better economic benefit and flexibility, as user can gain computing resources in demand, and releasing them when they are not needed. But when I/O bottleneck becomes a larger problem, all the benefits it brings fade away.The cause of this bottleneck is when the density of virtual machine on a physical server gets larger, more I/O traffic will be put on system bus, disks and network, which is a huge burden, especially when users run a lot of I/O intensive tasks on virtual machines.This paper is based on data center for laas private cloud of a flat network topology, and research on how to improve the I/O bottleneck situation by using prediction technology based on historical data and doing load balancing for both physical machines and switches connected to them, migration cost of virtual machines are also taken into consideration.