Research On Performance Optimization For Live Migration Of Virtual Machines

As a significant feature provided by the virtualization technology,live migration of virtual machines(VM)is a useful and important tool for the management of data centers and virtual cluster systems.It is widely used for the purposes of load balancing,power management,live system maintenance,high availability,etc.The researchers in academy and industry are continuously paying close attention to the technique of live VM migration because of its valuable effects.Nowadays,the need to dynamically deploy system resources is increasing in the environment of cloud computing.Meanwhile,the management activities in data centers are becoming more and more refined.So the operations of VM migration are becoming more and more frequent.Besides,the resource allocations of VMs are gradually increasing,and the VMs may carry out memory write-intensive workloads,which are difficult to be migrated.These trends all propose a higher requirement for the design of live VM migration system.In order to solve these problems,perform elaborate research work to improve the performance of live VM migration,including shortening the total migration time,reducing the amount of data transferred during migration,alleviating the application performance degradation caused by migration,etc.The research work is conducted from two aspects,i.e.,how to migrate the VM memory data and how to compress the data transferred with data compression techniques.In order to shorten the migration time,propose Bandwidth-Aware Compression(BAC)for efficient live VM migration.The compression techniques have been widely used for VM migration in order to reduce the data transferred and shorten the migration time.The experiments show that,the migration performance improvements obtained from different compression approaches are differentiated,and the improvements vary with the network bandwidth.So BAC monitors the network bandwidth available for the migration process,and dynamically chooses the suitable compression approach for VM migration according to the network bandwidth.Through the analysis of the memory contents of VMs carrying out various workloads,discover the similarity of different workloads when choosing the most suitable compression approach for VM migration.Thus design a list of compression strategies for BAC,and show that it applies to various workloads.Besides,the compression window sizes used in most compression algorithms are typically much larger than a single page size,so BAC employs multi-page compression to obtain more migration performance improvement.Experiment results show that,compared with the conventional methods which leverage fixed compression approaches for VM migration,BAC shortens the total migration time significantly,while achieving comparable performance for the total data transferred and the downtime.In order to reduce the amount of data transferred during migration and shorten the migration time,propose Multi-Frequency Memory Transferring(MFMT)for live VM migration.To achieve a shorter migration downtime,the pre-copy algorithm needs to transfer the VM memory with iterative process.The experiments show that,the dirtying frequencies of different memory pages may be differentiated a lot.The design in pre-copy algorithm to transfer all the dirty pages generated in one iteration again in the next iteration could lead to poor migration performance,because the pages with high dirtying frequency are often transferred many times during migration.This not only leads to the consumption of plenty of network resources,but also may impede the accomplishment of migration.In order to handle this problem,MFMT periodically obtains the dirtying frequency information of VM memory pages,and classifies the pages according to their dirtying frequencies.During VM migration,MFMT migrates the pages of different categories with different transferring frequencies.Besides,for the pages whose dirtying frequencies are higher than a threshold,MFMT only transfers them during the stop-copy phase.Thus MFMT accelerates the migration process.Experiment results show that,although the downtime is prolonged a little,MFMT significantly reduces the data transferred during migration and shortens the migration time a lot.Besides,MFMT could be combined with BAC,so that the combined optimizations would obtain more performance improvement for live VM migration.It is usually an intractable problem to migrate the VMs running memory write-intensive workloads,because they rapidly generate plenty of dirty pages.To solve this problem,propose the method to avoid transferring fake dirty pages along with the heuristic hybrid migration.Through a thorough analysis of VM dirty pages,obtain a novel observation: there are“fake dirty pages”existing during the pre-copy processes.Within the eight memory writeintensive workloads analyzed,four workloads have a proportion of fake dirty pages higher than 50% within all the dirty pages generated during migration.The content of these fake dirty pages is totally the same with the copies which have been transferred to the destination node in previous iterations.So the transfer of fake dirty pages is unnecessary,while wasting plenty of network resources.After exploring the reasons for the generation of fake dirty pages in detail,propose the method to compute and compare the secure hashes of memory pages to avoid transferring them.The collision probability of secure hashes is also discussed to guarantee the security of this method.Besides,in order to guarantee the success of migration for all the memory write-intensive workloads,further combine the secure hash method with the heuristic hybrid migration.The proposed hybrid migration scheme leverages heuristic and automatically switches from pre-copy phase to post-copy phase at the near-optimal moment.On one hand,this scheme avoids performing too many useless pre-copy iterations;on the other hand,it shortens the duration of post-copy phase as much as possible,in order to alleviate the negative impact on migration caused by post-copy's inherent weaknesses.Experiment results show that,the proposed scheme efficiently solves the problem of migrating memory write-intensive workloads and improves their migration performance significantly.The proposed scheme could be further combined with BAC,so that their optimization effects would together achieve more performance improvement for the migration of memory write-intensive workloads.