Research On Virtual Machine State Migration And Phase-Change Memory Wear-Leveling Method

System virtualization allows the sharing of the underlying physical machine resourcesbetween a multiple of virtual machines, which can raise the utilization and agility of computerresources. Phase-Change Memory (PCM) employs reversible change in electrical resistivityupon a change in material phase to store data, which is considered a promising storage tech-nique in the future. The resource management in virtualized datacenters depends on efcientand fast live migration of virtual machines. However, how to reduce the total migration timeand downtime in live migration at the mean time decreasing transferred data is still unsolved;running virtual machine can fail, but how to protect the virtual machine from failing faults,allowing the operating system, applications, and data to be continuously available, need moreeforts; PCM can only endure limited number of writes to its storage cells, how to avoid somestorage cells from being worn out sooner than the others to increase the lifetime of the wholedevice while reducing write overhead requires careful design. Since current solutions cannotanswer the above questions, we studied them and made the following contributions:(1) We present Microwiper, an improvement of memory propagation in live migration onthe basis of memory rewriting rate. Microwiper exploits two strategies: in ordered propaga-tion, dirty memory pages are transferred according to their rewriting rates; in transfer throttle,we factor available network bandwidth estimation in dirty page transfer to throttle copyingand sending dirty pages with the highest rewriting rate; after the accumulated rewriting rateexceeds the estimated bandwidth, next iteration is started immediately. The combination ofboth novel strategies can not only reduce dirtied pages by more than60%, but also shortenservice downtime and total migration time.(2) We design and implement Paratus, a failover system for virtual machines. Paratus isa novel framework that encapsulates operating system and applications in a virtual machine,and frequently replicates its state to a backup virtual machine periodically. In case the pro-tected virtual machine fails, the backup virtual machine can be instantaneously and seamlesslystarted to service the unfinished work of the failed virtual machine without user's perception.(3) We propose matrix wear-leveling for PCM. Our method views the storage blocks as amatrix and then remaps the logical addresses to diferent physical addresses to level the writesin every row and every column. Each logical address can be simultaneously remapped in a two-dimensional scheme, thus can be efciently remapped to any physical address efciently.The matrix wear-leveling can extend the lifetime of PCM both under normal applications andin malicious attacks while incurring very little write overhead.(4) We propose multi-way wear-leveling (MWWL) for PCM. MWWL levels writesacross blocks from a multiple of locations simultaneously and guarantees that each blockhas the potential to be remapped to the entire physical region for sufcient leveling. MWWLonly employs one-level address mapping, but can reduce access latency and hardware over-head, increase the endurance and security of PCM while incurring very low write overheadcompared with existing methods, which can efectively improve the lifetime of PCM.