Research On Adaptive Prefetching Strategy In The Multi-application Environment

Growth rate of the storage system performance is far below the one of the processor, and the performance gap has seriously affected overall performance. Cache and Prefetching can improve the performance effectively. Prefetching is a very hot researching topic, and many prefetching algorithms were raised by institutions all over the world. Generally, prefetching algorithms can be classified into three types: informed algorithm; sequential algorithm; access patterns algorithm.Sequential prefetching policy has good scalability, high predicting accuracy and is easy to implement.Most of current prefetching policies storage prefetched data and hot-spot data in a unified cache.Caching the two different types of data into different caches ,which are designed independently,can be more flexible. Being different to many algorithms, sequential detection module implemented in a separate data structure can record and use I/O information more effectively.The Multi-level hash table detection module can accelerate the detection speed and reduce the collision with limited space.A cache pool was designed to cache the prefetched data. The cache pool is composed of several cache blocks with various capacities, so as to cache prefetched data with suitable memory. LRAU(Least Recently Actively Used Algorithm),which is a variation of the famous LRU(Least Recently Used Algorithm)algorithm, is a more suitable replacement strategy for prefetching Cache.iSCSI and RAID(Redundant Array of Independent Disks) technologies are popular nowadays. iSCSI Enterprise Target is an open source software for iSCSI storage system. The prefetching policy was implemented in the blockio level.The result of the experimental shows that the adaptive algorithm can significantly improve the performance of the sequential I/O in the multi-stream environment.