Reducing write amplification of DM-SMR disks by shingle-aware persistent cache.

Area of Science:

• Computer Science
• Data Storage Technologies

Background:

• Shingled Magnetic Recording (SMR) technology increases data density by overlapping tracks.
• Track overlapping prevents in-place updates, leading to data corruption risks and poor random write performance.
• Drive-Managed SMR (DM-SMR) devices use persistent caches to mitigate these issues by batching writes.

Purpose of the Study:

• To introduce a novel, shingle-aware persistent cache cleaning policy for DM-SMR drives.
• To enhance random write performance and data integrity in SMR devices.

Main Methods:

• Developed a cache cleaning policy that merges cached updates by flushing writes in the shingle direction.
• Implemented a delayed write-back strategy to proactively free cache space by identifying safely writable data blocks and their optimal write order.
• Evaluated the policy using diverse block I/O traces.

Main Results:

• The proposed policy significantly improves cache management efficiency.
• Achieved up to 2.31X better performance compared to traditional set-associative baseline policies.
• Demonstrated effective freeing of cache space without incurring read-modify-write costs.

Conclusions:

• The shingle-aware persistent cache cleaning policy offers a substantial performance improvement for DM-SMR drives.
• This approach addresses the limitations of in-place updates in SMR technology.
• The delayed write-back strategy enhances the efficiency of data persistence in SMR systems.