Surbhi Palande and Puneet Mehrotra, University of British Columbia; Alexandra (Sasha) Fedorova, University of British Columbia and MongoDB; Margo Seltzer, University of British Columbia
Shingled Magnetic Recording (SMR) drives offer higher density and lower cost than Conventional Magnetic Recording (CMR) drives but are limited to sequential writes. To support random I/O, SMR drives require a Shingled Translation Layer (STL) located either in the device firmware (Device-Managed) or the host OS (Host-Managed). While anecdotal evidence suggests Device-Managed SMR drives exhibit unacceptable performance, leading to the misconception that the shingled medium is inherently inferior, we show that the root cause is the resource-constrained design of on-device STLs. We demonstrate that moving the STL to the host OS makes SMR drives competitive with CMR without requiring application or filesystem changes. Our host-based STL improves 99th+ percentile tail latency by ~40× compared to on-device implementations. In RAID reconstruction workloads previously cited as evidence of SMR failure, our solution achieves a 1.4× performance gain over CMR drives and a ~11× improvement over Device-Managed SMR.
Open Access Media
USENIX is committed to Open Access to the research presented at our events. Papers and proceedings are freely available to everyone once the event begins. Any video, audio, and/or slides that are posted after the event are also free and open to everyone. Support USENIX and our commitment to Open Access.
