Design And Implementation Of A File System Data Organization Method Based On Stripe Allocation Of RAID

On the arrival of Big-data Era,data in the Internet is growing at an exponential speed.Massive data in the Internet brings a serious challenge to exsiting storage techonologies.RAID is widely used in enterprise servers,and the application of SSD RAID can improve storage performance and degrade response time.Thus while adopting SSD to provide better hardware support,we can use RAID to achieve high performance read/write and provide data redundancy.However,the write amplification in RAID not only influences storage performance,but also hurt the lifespan of SSDs due to frequent parity updates.To address the problem,we propose a method of file system data organization called RAIDFS based on stripe allocation of RAID and implement it based on Ext4 file system.Our proposal designs a RAID-friendly data layout and cluster metadata to mitigate the write amplification due to metadata updates;We use a stripe buffer method,by modifying the mapping of file system,we can accumulate data in stripe units to improve full-write and large-write rate in RAID;To make the best use of buffer,we propose a space allocation algorithm of stripe granularity to manage stripe efficiently.We evaluate our file system with other state-of-the-art file systems under various workloads of Filebench and Postmark.The experimental results show that our file system is 10%-76% higher in throughput than Ext4 and F2 FS,and 10%-49% lower in response time,which verifies the feasibility and availability of our design and implementation.