Research Of Workload Aware Optimal FTL Algorithm For 3D Charge Trap Flash

With the rapid development of computer science,the data amount all over the world has grown explosively,the demand for storage capacity and performance is growing day by day.The gap between processor performance and bandwidth of the storage system is becoming bigger and bigger,and the storage wall seems to become a serious problem.Flash based storage system have become the mainstream storage device because of their higher performance,higher reliability,and lower power consumption then other storage devices.With the improvement of the flash manufacturing process,2D flash seemed to have a bottleneck in increasing flash memory capacity.The 3D flash uses tridimensional stacking technology to organize flash cell,breaking the bottleneck that traditional 2D flash has in increasing the capacity of flash memory.Thus leading a trend to become the new generation of flash in the future.When using traditional 2D flash,the atomic erase operation that takes a long time to complete blocks the read request,and results in the degradation of the reading performance of flash system.The 3D flash can take use of erase suspend/resume advanced command to alleviate the read request blocking problem.But there is another trouble caused by the latency of status switch,and thus frequent erase suspend/resume cannot optimize the read performance effectively.To solve the problem caused by frequently suspend,this paper propose a K-means based workload aware suspension relieve optimal flash translation layer algorithm named WASRFTL,which using K-means clustering algorithm to online partition the data into two type based on their hotness,and also use the characteristic caused by the one shot read/program of 3D Charge Trap stacking flash,pre-fetch the hot data into data read buffer in advance.So the efficiency of the data read buffer is better and amount of erase suspend/resume operation can be decreased,the flash memory system's reading performance can be further improved.The algorithm was evaluated by a trace-driven 3D flash simulation system.Under a series of real-world trace,the experimental results show that compared to the no-limited usage of erase suspension,the WASRFTL reduces the number of erase suspend/resume operations by 22% in average under all trace,increase the hit ratio of the data read buffer by 32.2% averagely,and the read request latency is reduced by an average of 16.4%.