Keynote 1: Challenges of Virtualized System: Performance Point of View by Hai Jin

Virtualization is a rapidly evolving technology that provides a range of benefits to computing systems, such as improved resource utilization and management, application isolation and portability, and system reliability. Among these features, live migration, resources management (including vCPU scheduling and I/O management) are core functions. Live migration of virtual machine (VM) provides a significant benefit for virtual server mobility without disrupting service. It has become an extremely powerful tool for system management in a variety of key scenarios, such as VM load balancing, fault tolerance, power management and other applications. Experimentations and traces show that the performance of live migration is not good enough for different applications. Of course, based on the virtualization architecture management schemes for CPU and I/O resources also need to be reconsidered when supporting different applications with different workload. In this talk, some typical issues in virtualized system will be discussed. Firstly, to take into account the migration overhead in migration decision making, we thoroughly analyze the key parameters that affect the migration cost from theory to practice, and construct two application oblivious models for the cost prediction by using learned knowledge about the workloads at the hypervisor (also called VMM) level. We evaluate the models using five representative workloads on a Xen virtualized environment. Based on the model, we have proposed two live migration schemes for different scenarios: a novel approach that adopts checkpointing/recovery and trace/replay technology to provide fast, transparent VM migration for applications with high reliability, and a memory-compression-based VM migration system for normal applications. Secondly, the asynchronous-synchronous disk I/O model in a typical virtualized system exhibits several problems. For example, when the frontend fails abruptly, the unsaved data in the frontend´s cache - ill be lost. To address the problems, we introduce a new I/O model. In this model, rather than performing the asynchronous-synchronous operations for an asynchronous I/O write request, the frontend file system uses synchronous operations to deal with the I/O request and the backend file system performs asynchronous operations to write the data to the hard disk. A prototype system called HypeGear is implemented on the Xen hypervisor. Thirdly, VMM schedulers have focused on fairly sharing the processor resources among domains, rarely consider VCPUs´ behaviors. However, this can result in poor application performance to overcommitted domains if there are concurrent programs hosted in them. We review the properties of both Xen´s Credit and SEDF schedulers, and show how these schedulers may seriously impact the performance of the communication-intensive and I/O-intensive concurrent applications in overcommitted domains. A novel approach, that dynamically scales the context switching-frequency by selecting variable time slices according to VCPUs behaviors, is then proposed to improve the Credit scheduler more adaptive for concurrent applications.