TLC NAND could penetrate biz with flash-to-flash backup
Who'll lead the dash to flashedy-flash?
VMworld Barcelona Violin co-founder and CTO Jon Bennett identified a potential enterprise role for three-level cell (TLC) flash at VMworld Barcelona today: flash-to-flash backup. TLC is currently seen as pure consumer technology.
The first TLC NAND SSD has been announced by Samsung. But TLC NAND has a short working life, with 500 - 750 phase/erase cycles, rendering it unfit for enterprise flash storage which is reckoned to need program-erase cycles of 2,500 and up at least.
Bennett suggested that TLC flash could be used for backup, not because it would shorten the backup window but because its restore time would be very fast compared to disk-based backup systems.
He thinks that enterprises backup changed data and that is around 5 to 10 per cent of system data written to backups per day. He calculates that this could equate to thirty full device writes a year and, with a 1,000 PE cycle TLC product, that would give you three to four years of working life; good enough.
Bennett reminds us the enterprises use D2D (disk-to-disk) backup systems mostly for the vastly faster restore time compared to tape.
The short working life wouldn't matter because the write rates would be low: "With just 5 to 10 per cent of system data backed up per day then TLC endurance is okay. … Backup time to TLC should be shorter, but restore rate is where it matters the most. Restore is random, unlike (backup) ingest and that's where the real benefit would come. Shrinking restore time from the hours of disk systems to minutes could have real benefits for the enterprise."
Could we see a backup TLC flash tier inside existing flash arrays? "You would have backup to flash products. If it's backup to flash in the array it's not a backup, it's in the same array – it's not protected." Basically, if the array goes down, the backup goes down with it.
Bennett said: "We may also see TLC used in a nearline storage mode." He emphasised that these are technology visions and there is no commitment by Violin Technology to use TLC flash or have TLC product on its roadmap. This is just a CTO thinking about technology.
Still, F2F, SLC and MLC flash to TLC flash backup – it has a ring to it. Maybe there's a flash-focused start-up somewhere thinking Data Domain 2 is on the way. ®
Re: data for movies
Movies range up to 250GB for the 3D format. They use a very high resolution 4K graphics image. The result is they don't fit Blu-Ray disks, and the use of multiple disks would be a nightmare to manage. Currently movies go out on HDDs in special rugged cases. If the HDD were replaced with SSD, the cost would probably be close, since the cases are very expensive.
They are typically shipped via common carrier, and HDDs do get damaged. This can be a potential hit for theaters since only one copy is shipped which runs from a central server. Likely SSD would be more reliable.
Return of the Data
At the current state-of-the-art, the idea of yet another storage tier using TLC is worth discussion, but it isn't a feasible play yet. Any backup device has to guarantee long-term retention of the data, and a low error rate on that data when it is restored. This is especially true of highly de-duplicated data.
Looking at restore, there are a couple of modes involved. Restoring a single file is a random access event. While flash will be faster, the overhead of loading programs, selecting files etc will swamp the speed difference. In batch restores, the issue is sequential throughput, where flash has a much smaller edge over HDD. Here, the cost per GB of HDD is so low compared with flash that extra drives are affordable, and the backup array can thus be lifted up to flash levels and still be cheaper.
Flash backup isn't a near-term proposition, except perhaps in Mil-spec applications. Ultra-rugged HDD modules for distributing movies to theaters could be replaced by flash, but the capacities involved probably don't require TLC.
Incremental backups on unreliable media. What could possibly go wrong?