Any further dramatic improvements to the capacity of hard disk drives is bound to come with some major changes to the currently employed techniques, as they are reaching their limitations imposed by the laws of physics. Of the new technologies being explored, only Shingled Magnetic Recording (SMR) writing promises a high areal density increase with minimal changes to the manufacturing process.
SMR overlaps the data tracks on the disk surface like the shingles on a roof. A random write in a SMR disk will overwrite and destroy data on subsequent tracks, forcing the SMR disk to be a largely sequential write device with unrestricted random reads.
Although it is possible to treat a SMR disk like a virtual tape with random reads, exploring its potential to replace hard disk drives in their traditional roles is more desirable. We are exploring the SMR data management issues and are looking at solutions that enable effective utilization of SMR disks in a wide variety of applications.
Further development and assessment of data management solutions is hindered by the unavailability of SMR disks, as the hardware is still under development. To solve this problem, we developed a block-level pseudo-device driver to emulate a SMR disk on a regular hard disk drive. Since the underlying hardware of a SMR disk will be very similar to existing hard disk drives, the read/write performance of our emulated disk should closely match that of a real SMR disk.
We are currently working on effective data layout mechanisms, background garbage collection techniques and efficient metadata management solutions designed specifically for SMR disks. There are couple of different SMR interface proposals, calling for data management solutions that work either from the host-side, or the disk-side, or from both ends. We are developing solutions that benefit all cases.