0

In case you don't know, Backblaze is somewhat famous for making custom ginormous boxes of disks made entirely out of consumer grade parts for their file hosting business. Their newest version holds 135 terabytes for $7,384, which is significantly cheaper than any existing "enterprise" solution.

For reference, the "pod" holds 45 disks. If their 2.0 pod uses the same basic configuration as the 1.0 pod, every 15 disks is in its own RAID6 array with 2 pairity disks out of the 15. In total this leaves 87% usable space. Failure rate isn't too much of a problem; over their total 9,000+ hard drives, an average of 10 per week fail, or 5%. The newer 2.0 pods see less than 1% failure rate

However I thought that running consumer hardware as a server was bad. Especially on ServerFault, people get on to them like a pack of angry wolves saying "Don't use consumer hardware, use server hardware!" There was a giant answer I saw a while ago that talked about all the horrible things that result from using Consumer SATA disks (Failure rate, speed, RPMs, Bad Sector in Raid issues, not being "certified" etc). Here's another one. Sometimes they are called "toys". Mark wouldn't even put a consumer grade server ON the server's network.

With all the horrible issues with consumer hardware, how is Backblaze able to run their entire business from them? How was the Sysadmin able to sleep at night? Was all the hype about enterprise drives just fear mongering or blown out of proportion?

More importantly, why can't I or someone else do something similar? 135 terabytes seems like a lot of data for such a cheap price. Even two pods mirroring each other seems like it would be cheaper than the equivalent "enterprise" solution, and in some cases it might be more redundant since its essentially RAID-60 over two separate machines.

Thoughts?

Community
  • 1
TheLQ
  • 1,013

2 Answers2

5

Backblaze has a business model that allows them to do this. They need gobs of storage space (though it doesn't have to be particularly fast) and they need it cheaply. It's worth it for them to dote on storage because they need so much and they can scale their production to handle having sufficient spare parts and support labor on-site.

If your company's business model allows you the convenience of building your own hardware (and the curse of being the sole support provider when you have failures) then, by all means, go for it. For most companies it's cheaper to purchase "enterprise" solutions because of the technical support entitlement, compatibility testing and vendor certification, and warranty support that comes with an "enterprise" solution.

You're not going to get the performance out of "consumer" disks that you will with "enterprise" disks. If you're talking about large "near-line" storage needs that's one thing, but "enterprise" disks (magnetic or SSD) are necessary if you want to scale to large numbers of transactions (I/O operations per second) on an array of disks. SATA disks just aren't as fast as SAS disks.

If you have a business that needs massive amounts of raw storage and can handle being your own support provider and the expense of "rolling your own" then using the Backblaze designs might well work for you. I think you'll have to scale fairly large to see benefits as compared to just buying something off-the-shelf in terms of total cost of ownership and long-term reliability.

Evan Anderson
  • 142,957
4
  • lots of redundancy
    • IMO Consumer grade hardware is not inherently evil, but you should certainly understand what you are getting and plan properly for their failure and replacement. There is a big difference between the really low-end consumer gear, and the high-end server gear, there is also some stuff in the middle that could be perfectly fine in some production settings.
  • The are running a backup business
    • Which most likely means lots of nice long sequential writes, followed by lots of time where the drive is doing absolutely nothing, combined with the very rare read for recovery. This is certainly a drastically different usage pattern then you would see for an Mail/Database server.
  • Enterprise equipment is often focused on IOPS
    • Low end-gear is almost certainly cheaper, but people buying enterprise solutions are frequently buying IOPS, not capacity. The IOPS difference between a cheap device and a enterprise gear can be huge.
Zoredache
  • 133,737