Seesaw: Performance Guarantee for RAID Reconstruction

Today, the disk failures in large-scale storage can occur almost on a daily basis and the reconstruction process is very common. Additionally, the reconstruction process is always time-consuming. When RAID is in the reconstruction process, the performance of front-end application will be very poorly. This will significantly impact the end user´s feeling in storage system availability. In order to improve the storage system availability, we propose an architecture which controls the reconstruction speed and provides enough bandwidth to front-end application. Our approach performs well on the write-dominated trace and speeds the average response time up to 8 times.