大规模集群环境中虚拟机迁移关键技术及相关算法研究

Digital data explosion mandates the development of cloud cluster technology. Because of the ease with which computing and storage resources can be wrapped and migrated around the cloud cluster, virtual-machine migration technology becomes more and more important in order to transform the cloud cluster from a collection of heterogeneous physical nodes to a huge pool of homogeneous resources. Unfortunately, none of existing approaches and technologies can provide optimized support for seamless live migrations of virtual-machines in cloud cluster, let alone the vision of efficiently pooling geographically distributed cluster resources. Furthermore, new challenges arise in this field, such as the problem of parallel scheduling of virtual-machines migrations and dynamic consolidating of resources in cloud cluster..In this proposal, we plan to present the non-suspend live migration approach for the first time. With the needs of research on multi-mode & multi-granularity shadow memory mechanism, storage trace-map mechanism, user-mode migration protocol stack, transfer block based access pattern prediction technology, and the non-suspend data scheduling model & algorithm, the non-suspend live migration approach can support seamless transfer of virtual-machines over low bandwidth and high latency links while minimizing the performance degradation of migrated services and applications. We also plan to propose a cloud migration framework which provides a solution to find the optimized algorithm and strategy for parallel migration scheduling problem in cloud cluster.

数据的裂变式增长，使得面向海量数据处理的大规模集群技术成为了未来的研究方向，在这样的背景下，虚拟机迁移技术能够打包资源与服务使其跨越物理节点边界的鸿沟，将异构的集群硬件资源转换为一个同质资源池，其重要性不言而喻。但是，已有的迁移技术都存在不足，面对不断扩张的集群规模无法做到一致高效的实时迁移，同时，新环境也带来了新挑战，如多虚拟机的并行迁移调度问题。本项目拟首次提出无宕机实时迁移技术，以多模式可变粒度Shadow Memory、外存访问跟踪、堆栈式用户态迁移体系等技术和机制为支撑，以最小化迁移时间、最大化服务质量为目标，结合内外存数据调度模型与算法，实现大规模集群环境中的虚拟机一致高效实时迁移。同时，本项目拟对迁移框架技术展开深入研究，通过行为模拟和成本估算，将多种虚拟机迁移过程模型化，进而以成本最小化为目标，设计并实现一整套指导和协调大规模集群环境中虚拟机并行迁移的相关算法。

数据的裂变式增长，使得面向海量数据处理的大规模集群技术成为了未来的研究方向，在这样的背景下，虚拟机迁移技术能够打包资源与服务使其跨越物理节点边界的鸿沟，将异构的集群硬件资源转换为一个同质资源池，其重要性不言而喻。但是，已有的迁移技术都存在不足，面对不断扩张的集群规模无法做到一致高效的实时迁移，同时，体系结构的发展，如事务内存的出现，大数据处理浪潮的兴起，也给大规模集群环境中虚拟机的迁移、管理及应用带来了新挑战。本项目拟首次提出无宕机实时迁移技术，以可变粒度Shadow Memory、外存访问跟踪、事务内存处理、堆栈式用户态迁移体系等技术和机制为支撑，以最小化迁移时间、最大化服务质量为目标，结合内外存数据调度模型与算法，实现大规模集群环境中的虚拟机一致高效实时迁移。同时，本项目拟针对大数据处理的场景，探索虚拟化迁移技术的应用框架，研究结合虚拟化迁移技术、虚拟容器存储技术、面向底层平台的优化技术的云化大数据处理平台及其相关算法。