VMware: Storage data functions can and will migrate to flash media
Speed and simplicity
Comment VMware's CTO has revealed his belief that array and VSAN-level storage data service functions can, and will, move into flash media.
He made the point on a recent blog that it is logical for physical block addressing to be dynamic rather than static.
A disk drive's firmware maps logical block addresses (LBAs) to physical block addresses (PBAs), and this is a static relationship.
In a flash drive, because of the need to write and erase at the block level, we have out-of-place updates, the LBA-PBA relationship is dynamic and the drive’s flash translation layer maintains the equivalent of a log-based file system.
VMware’s Sandeep Uttamchandani, a developer of next-generation storage products, blogs that based on this realisation, flash media devices can carry out functions currently performed by a disk array controller (or VSAN controller entity).
- Integrity checksums
He sees this happening in three phases, which could overlap:
- Existing block IO standard extensions
- IO management plane integration
- Next-generation data management interfaces
Some of this can apply to disk drives as well.
Uttamchandani said he expects we will see application-centric interfaces such as key:value stores, and of course we have already seen the first example of these with directly addressed Ethernet drives from Seagate and HGST.
Another early example was Fusion-io’s work with Atomic Writes.
The benefits of this migration of storage data services into flash media are said to be the speeding up of server applications, the ability to cram more applications into servers and storage infrastructure simplification.
For example, if VMware VSANs can offload such data services to the flash storage media, that would leave more server CPU resources for running virtual machines. ®