Techniques Mitigating Update-Induced Latency Overhead in Shingled Magnetic Recording

Shingled writing has recently emerged as a promising candidate to sustain the historical growth of magnetic recording storage areal density. However, since the convenient update-in-place feature is no longer available in shingled recording, in order to update one sector, many sectors must be read and written back, leading to a significant update-induced latency overhead. This work develops two simple design techniques that can reduce such a latency overhead. Because the spatial locality of update-invoked read operations naturally enables the use of the 2-D read channel signal processing, the first technique aims to reduce update-invoked read latency by trading the SNR gain obtained by a 2-D read channel for higher disk rotation speed. Since update-induced latency overhead strongly depends on the location of the sectors being updated within each shingled region, the second technique aims to reduce the latency overhead by leveraging the data access locality in most real-time workloads in order to determine appropriate data placement. Through extensive simulations, we show that disk rotation speed boost assisted by a 2-D read channel can reduce the update latency by up to 33%, and data access characteristic sector placement can reduce the update latency by over one order of magnitude.