Optimal Write Batch Size in Log-Structured File Systems

Log-structured file systems create a performance benefit by maximizing the rate at which disk write operations can occur. However, this benefit comes at the expense of individual read operations, which can experience long queuing delays due to batched write service. This paper presents an analytic derivation of the write batch size that minimizes read response time, while retaining the benefit of write batching. Simulations that relax the assumptions of the analytic model demonstrate that the analytic result can be used to advantage in a practical setting.