Reseach On Non-Volatile Memory Optimization For Building High Performance Mobile Devices

Thanks to the advances in mobile microprocessors and operating systems,the performance of mobile devices,such as smartphones,tablets have been significantly improved.Various feature-rich applications developed for mobile devices bring unprecedented user experience.However,this comes with a prices: the richer functionality an application can provide,the more demanding it is on computing and memory resources.Nevertheless,due to the high energy consumption and poor scalability,the dynamic random access memory(DRAM)has become one of the performance bottlenecks for mobile devices,degrading the user experience.Therefore,new main memory technology are urgently demanded by mobile devices.The emerging non-volatile random access memory(NVRAM),such as phase change memory(PCM),spin-transfer torque RAM(STT-RAM),and resistive RAM(RRAM)has been considered as one of the most promising candidate for replacing DRAM because of its nice features such as non-volatility,low access latency,low power consumption,high density and high scalability.Therefore,this thesis target at improving the performance and user experience of mobile device leveraging the emerging NVRAM.First,we propose NVRAM-based swap area to improve the performance and reduce the power consumption of mobile devices.Then,wear-leveling algorithm and write reduction mechanism are designed to prolong the lifetime of NVRAM swap area.Finally,NVRAM-based optimization techniques are proposed to achieve fast and lightweight virtual machine(VM)checkpointing in virtualized mobile devices.The key contributions of this thesis are summarized in below.(1)To solve the performance issues of flash-based swap area,we revisit swapping for mobile devices with NVRAM technologies.Instead of using flash,we build the swap area with fast and byte-addressable NVRAM,to allow high performance and low power consumption.Based on NVRAM's high performance and byte-addressability,we show that a copy-on-write swap-in scheme can avoid unnecessary swap-in operations by reading the pages in NVRAM swap area directly without copying them back to DRAM.Since NVRAM has limited write endurance,to avoid fast worn-out of certain NVRAM cells,we also propose Heap-Wear,a wear leveling algorithm that can evenly distributes writes across the whole NVRAM swap area.(2)NVRAM has limited write endurance and the current victim page selecting algorithm does not aware it.Therefore,the design of an NVRAM-based swapping system must also consider the endurance of NVRAM.In this thesis,we propose a NVRAM-aware swapping subsystem,which prolongs the lifetime of NVRAM swap area in mobile devices by reducing the write activities to swap area.Different from traditional wisdom,such as wear leveling and hot/cold data identification,we exploit the facts that code pages are easy to identify,read-only,and therefore a perfect candidate for swapping.Utilizing NVRAM's byte-addressability,we support execute-in-place(XIP)of the code pages in the swap area,without copying them back to DRAM based main memory.(3)Conventional VM checkpointing involves dumping the VM's entire memory to storage,which induces a significant amount of(slow)I/O operations,degrading system performance.In this thesis,we propose to optimize VM checkpointing by taking the advantage of byte-addressable NVRAM.In this thesis,we propose a working set pages identification method to divide VM's memory into working set pages and non-working set pages.Utilizing the non-volatility of NVRAM,we store working set pages in NVRAM to reduce the I/O operations and achieve fast VM checkpointing.To further reduce the write activities to storage,we propose an energy-ware data deduplication technique to eliminate redundant data in VM snapshot and save storage space.We evaluate the propose techniques in real mobile devices.The evaluation results suggest that the proposed NVRAM-based swapping can effectively improve the performance and reduce the energy consumption of mobile devices.Moreover,the writes to NVRAM are evenly distributed and reduced.The proposed VM checkpointing optimization techniques can also reduce the I/O operations and improve the performance of checkpointing in virtualized mobile devices.