Research On Storage Cluster Hierarchical Architecture-aware Locally Repairable Codes

In recent years,erasure codes technology has been widely used in distributed storage systems due to its low storage overhead and high fault tolerance.The high repair cost of the erasure codes not only reduces the data recovery efficiency,but also affects the overall performance of the storage system.LRC code utilizes additional storage overhead to reduce single-node repair cost,but because of its flat structure,it still consumes a large amount of scarce inter-rack repair bandwidth,which limits its improvement of repair performance.To improve LRC for the hierarchical architecture of the storage cluster,this study do the following works: The layered LRC(LLRC)is proposed: the LLRC framework using hierarchical placement and layered recovery is given,which places multiple coding blocks into one rack and divides the repair process into intra-rack and inter-rack.Two LLRC codes based on LLRC framework are proposed,and corresponding placement and repair methods.And an explicit construction of the underlying LRC code is given.The character of LLRC is theoretically analyzed: the repair cost analysis shows that LLRC can greatly reduce the consumption of the inter-rack repair bandwidth;the code fault tolerance analysis shows that LLRC has considerable storage fault tolerance efficiency;the system reliability analysis shows that LLRC achieves higher reliability due to its high repair efficiency;cost-benefit analysis shows that LLRC achieves better trade-off between storage overhead and inter-rack repair bandwidth cost.Then an LLRC protype storage system is designed and implemented to simulate the I/O and code/decode process in a real distributed storage system.In this system,a variety of coding strategies such as LLRC codes are implemented,which shows the application of LLRC.The test results of the LLRC's show that the inter-rack network transmission does limit the repair performance of the coding storage system and LLRC codes greatly improves the repair performance by reducing the inter-rack repair bandwidth,which reduces degraded read time by up to 69.5% and increases node recovery efficiency by up to 292.6% compared with the LRC code.