虚拟化网络性能评测和优化技术的研究.pptVIP

Virtualization is a key tool in the data centre, because it enables multiple virtual machines (VMs) with operating systems and applications to run within a physical server. With the data growing fast and applications becoming more disk-intensive, allocating disk resources efficiently while preserving a high disk throughput becomes of key importance in virtualized environments. As we all know, Exploiting spatial locality is an important technique for improving disk efficiency and performance. For example, traditional file systems often allocate the accessed data of a process as contiguous blocks if possible, so disk scheduling can easily exploit spatial locality to reduce disk seek and rotational overheads. But achieving high spatial locality is a challenging task in a virtualized environment. * Why is it difficult to exploit spatial locality ? On one hand, VMs as whole running environments encapsulate more than one process, such as virtual desktop, data intensive application with parallel processes accessing data. On the other hand, the transparency feature of virtualization causes semantic isolation between the hypervisor and guest VMs. As a result, the block I/O layer lacks a global view of the I/O access patterns of processes. * A huge number of studies have been dedicated to improving the performance of disk-intensive applications in virtualized environments, On one hand, some of these solutions use invasive mode scheduling to manage I/O requests, such as Selecting the disk scheduler pair within both the hypervisor and VMs according to access pattern of applications, but it introduces an additional Hypervisor-to-VM interference. On the other hand, some solutions use non-invasive mode scheduling to manage the I/O requests in virtualized environments without destroying the transparency feature of virtualization, such as streaming scheduling and antfarm. But they can only infer read streams, and cannot be applied for write applications or when mixed appli