October 2nd, 2009

priate fish

We are facing an impending storage explosion at work

The tidal wave of schools wanting home drives and email accounts for their students is about to break over our heads. I have been talking options with my team -- our discussions have ranged from high-end solutions to Frankenstorage to everything in between.

It just amazes me what the major storage vendors charge for their products. EMC is the highest-of-the-high-end, with NetApp (nee Network Appliance) between EMC and the likes of Dell and Sun. We have a NetApp SAN, with several shelves of 10K RPM 300 GB FibreChannel drives and a single shelf of SATA drives (for disk-to-disk-to-tape backups). A single TB of usable space cost us $9,460 with the FC shelves, and $4,892 for SATA. (That was after the super-aggressive introductory pricing wore off.) The kicker is that these costs are based on adding to our existing storage controllers. Adding a shelf once the capacity of the controllers is reached puts you into a whole new world of pain. Think $40K per usable TB.

I don't mean to offend the NetApp folks, but there's nothing overwhelmingly special about these disks. They're made by the same companies that make the disks that you and I buy. Yes, there are all sorts of wonderful features in the system ... but they'll all cost you extra to "unlock" and to maintain. It almost seems like the five star hotel effect. It's a five star hotel so you expect to pay a lot, even if the breakfast buffet isn't any different than the one at the Marriott down the block.

We've looked a assembling our own kit from commodity pieces, but there are always questions. Who will support the hardware? is a big one. BackBlaze -- an on-line backup vendor that needs staggeringly vast tracts of space -- has developed a nifty commodity box that holds 65 TB of disk in 5 rack units. Very attractive, until you understand that their givens are not your givens. It's the Google problem -- they too operate on dirt-cheap commodity hardware, and they too created their entire operating environment from the ground up. The designed each cheap server node to be expendable. In essence, they have data redundancy at the service level so they don't have to pay for redundancy at the device level.

What's a medium-ish sized IT shop to do? Piecing together an infrastructure from parts may work for a startup, but we can't do that here. Windows servers don't fit the Google/BackBlaze model, and "enterprise class" storage from EMC or NetApp will break the bank. The answer may be a middle path. Taking a good Dell server (R710, loaded with RAM, a 10Gb Ethernet card and a monster RAID card) and pairing it with one of their external drive enclosures yields costs in the range of $4,400/TB for high-speed storage and $1,300/TB for archival or D2D2T storage ... even with the gold-plated 3-year 24x7 4 hour response support plan. Linux is a perfectly mature iSCSI target / NFS server, so I know that we can make it work.

Film at 11. We'll see how this turns out.
  • Current Music
    Miami Vice - the complete soundtrack
  • Tags
    , ,