ExaGrid makes disk-based purpose-built backup appliances and its CEO told us he sees no need to move to flash media.
We suggested ExaGrid might add an SSD front end as a competitive response to flash-based backup target systems from Pure Storage, StorOne and VAST Data, not forgetting the DRAM-cached InfiniGuard system from Infinidat.
An email conversation then ensued between ExaGid CEO Bill Andrews and B&F in which he pretty much squashed the proposition.
Blocks & Files: Why wouldn’t you use flash media in the front end backup job Landing Zone? It’s faster than disk.
Bill Andrews: Standard primary storage disk is slower as it is typically scale-up and CIFs. ExaGrid uses the proprietary protocols which are faster such as the Veeam Data Mover, Veritas NetBackup OST, etc. This is true of flash as well.
But ExaGrid has done the math. To put flash in the Landing Zone would greatly increase the cost and resulting price without the added value. The backup performance is no different. First the media servers cannot keep up so the flash would sit there idle. It is overkill.
Second, flash file systems are built for small database transactions and not large sequential files so, believe it or not, we beat flash in head to head tests because we are optimized for large sequential files and we also do job concurrency.
Blocks & Files: How is ExaGrid optimized?
Bill Andrews: ExaGrid is optimized for large backup jobs. In backup you have huge jobs that consist of hundreds to thousands of files per job. Other storage including flash is not. Disk is optimized for Word, Excel, etc. files. Flash is optimized for database transactions.
Blocks & Files: The key constraint then is the media server?
Bill Andrews: ExaGrid is scale-out and ExaGrid uses job concurrency. Typically, the media servers are not putting the data out fast enough to overwhelm ExaGrid and the cost to get media servers that would is too expensive. If this dynamic changes then we may do flash in the Landing Zone someday.
Blocks & Files: What’s the take away here?
Bill Andrews: The net is customers test ExaGrid against flash and say: “Wow you were faster and you are a pittance of the price.” We don’t get value from adding flash to our front-end because we don’t need any additional performance.
All of the above is why, at this point, we are faster than anything out there and being faster still may or may not be possible depending on the media servers, and even if it was, the question is: what will the customer pay for that? (cost versus value equation).
We think two other factors might influence things in the future. One is having the media servers get faster because they run in flash-based systems. The second would be a rising need for faster restores. Other things being equal a restore from SSDs ought to be faster than one from disks.