VSimilar:A VM Scheduler Enhancement For High Adaptability Based On The Cpupool Mechanism

Virtualization systems(e.g.Xen)provide powerful,flexible and scalable infrastructures for big data processing.In the virtualization system,the virtual machine(VM)scheduler is the key component to efficiently consolidate the multiple VMs on the virtual machine monitor(VMM),a.k.a.hypervisor.The default scheduler of Xen,the Credit scheduler,should assign the system resources to each VM adequately according to the high dynamic attributes of the VM tenants due to the unpredictable customer application behavior changes.However,the Credit scheduler still has the deficiencies in terms of adaptability for tasks of enormous variability and variety in the environment of big data.The Credit scheduler is not able to mitigate the I/O latency effectively for I/O-intensive VMs.Moreover,the Credit scheduler is not friendly to none-I/Ointensive VMs.In addition,the Credit scheduler shows low adaptability for VMs due to the same scheduling manner without distinguishing.VM classification and resource allocating specification are the mainstream idea to improve the shortcomings of the Credit scheduler.Among the approaches which attempt to improve the VM scheduler basing on this idea,the Xen organization official has chosen the cpupool mechanism and has integrated the solution to Xen.The core idea of the cpupool mechanism is to divide the physical CPU cores on the machine into different pools.Each of the pools has an entirely separated VM scheduler,and can be set with different scheduling parameters.However,the cpupool mechanism contains no algorithms to classify VMs or the time-slice allocation strategy,which are necessary for a complete implementation of a high-adaptive VM scheduler.In this paper,we discuss the optimizing opportunities and introduce a VM scheduler enhancement solution for high adaptability,called vSimilar basing on the cpupool mechanism,which uses the VM multi-classification model to optimize VMs running different tasks in a more VM dynamic suited manner.Moreover,by a dynamic time-slice tuning function,vSimilar manages to provide a more reasonable time-slice allocation strategy,which allocates resources to VMs of different running characters heterogeneously.Both the two enhancement measurements promise vSimilar sufficient adaptability of virtualization system for variable and various data processing tasks.The experimental evaluation shows that vSimilar can remarkably improve the performance of Xen virtualization system.For example,a VM server hosted by Xen with vSimilar can limit about 90% a client's Ping round-trip time(Ping RTT)in 5ms;vSimilar is proved to be pretty friendly none-I/O-intensive VMs;vSimilar also increases the page operation rate by nearly 17.8% for a Xen-hosted VM Web server.