Peter Eicher has a good article over at Computerworld entitled “Will We Ever Use Flash Drives for Enterprise Backup“. Peter bets that in a few years that flash arrays will commonly be used as backup targets. He discusses common objections to the idea: cost and limitation on read/write cycles.
I thought that it might be interesting to tackle this question from a backup appliance point of view – rather than a backup software only one. One advantage to working with physical backup appliances is that you get a pretty decent feel not only for backup software, but backup hardware (e.g., servers, storage, networking, etc.) as well.
Of course, Unitrends currently uses Flash drives to optimize its 100TB (in only 4Us of rack space!) backup appliance – the Recovery-943. This device uses a three-tiered architecture with mirrored SSDs (Solid State Drives) at the first tier, higher-speed drives at the second tier, and lower-speed drives at the third tier. The software embedded on the backup appliance optimizes the many functions of the Unitrends backup appliance – from backup to disaster recovery (replication) to instant recovery (failover virtualization) to archiving (D2D2x – or Disk-to-Disk-to-Any) to dissimilar bare metal to content-aware byte-level deduplication.
However, what Peter posits I believe is the concept of using Flash arrays as the primary backup storage area for data. I don’t share his views concerning this occurring in the next few years in a general case for large backup sets. And let me note that I say that reluctantly – a lot of thorny engineering problems go away if you assume much higher speed I/O.
The reason I don’t think it’s going to happen isn’t that cost of SSDs isn’t going to go down – that cost will decrease as Peter predicts. And it’s not because of the read/write cycle limitations – that will be addressed over time. The real reason I think that rotational (spinning) drive value, as measured in price per terabyte, will continue to fall as the SSD price per terabyte falls. And the capacity increases in rotational drives are dramatically increasing.
In any case, that’s my opinion – what do you think?