近年来，随着互联网规模的指数式增长，公司网站存储了越来越多的数据，使得用户请求的响应速度越来越慢。一种有效的解决方法是引入键值存储系统（KV Store）。KV Store将数据按照键值对的形式进行存储和索引。按存储介质的不同，KV Store分为基于内存的KV Store和基于磁盘的KV Store。按工作模式的不同，KV Store分为缓存模式的KV Store和存储模式的KV Store。本项目着眼工作于缓存模式的、基于内存的KV Store，主要研究三项关键技术：哈希表设计、内存管理及换入换出策略、并行加速技术。本项目拟提出两种哈希表设计方案、一套高效的、内存冗余少的内存管理方案和置换策略，然后在四个平台（多核CPU、众核、GPU、FPGA）上研究并行加速方案，最后拟实现多个平台的KV原型系统并进行大量实验测试，定位系统瓶颈，进一步优化算法和数据结构。

In recent years, with the exponential growth of Internet scale, there are more and more data stored in company websites, consequently, the response speed for user requests become slower and slower. One effective way is to introduce Key-value Store (KV Store) system, which trade data as key-value pairs for storage and retrieval. According to the storage medium, there are two kinds of KV Stores: the first kind is stored in RAM, and the second kind is stored in hard disk. According to the working mode, there are also two kinds of KV Stores: the first kind works in the cache mode, and the second works in the storage mode. This project focuses on the KV Stores that are stored in the RAM and work in the cache mode, and mainly studies three key techniques: hash table design, memory management & cache eviction policy, and parallel acceleration techniques. This project will propose two different designs for hash table and one set of solutions for efficient memory management & cache eviction policy. Then we will study parallel acceleration techniques on four platforms: multi-core CPU, many-core, GPU, FPGA. Last, we will implement KV Store systems based on multiple platforms, conduct extensive experiments to figure out the system bottleneck, and further improve the key algorithms and data structures.