The Gig of Ham

One geek's contributions to the series of tubes

Jan 21, 2015 - 3 minute read - Comments - sysadmin

On spinning rust

It’s that time again, Backblaze has released another report on what’s the best/worst drive. The information is interesting, but I have problems with the widespread “conclusions” and some of the statements made on value. Having worked in several large deployments of hard drives I have a major rebuttal to every Backblaze report:

You should not buy the cheapest hard drives possible for use in a Data Center.

It’s that simple. Don’t do it. Backblaze is a very, very specific use case and it doesn’t map well to just about anything other than their business segment (I’m betting Carbonite and CrashPlan have similar drive procurement, operation, and requirements). They buy the cheapest drives possible and don’t care when they fail or how much they cost to operate. Because of this, they tend to get bottom of the barrel consumer drives and those devices simply suck. One of my previous employers did this as well, and it was a nightmarish hell-scape of failed drives, failing drives, and failing RAID controllers. Just don’t do it. Spend the extra $20/drive and get NAS or Enterprise rated disks. You’ll be happier, and so will your operations team. (Also don’t use RAID controllers, but that’s a whole other discussion for some other time.)

Which brings us to Seagate and “value”. I intensely dislike Seagate drives, and have for years. They are cheap, failure prone, and hot. The first point is fine, the other two not so much. Seagate enterprise drives are better for reliability but they are still hot and frequently not cheap. Heat is the enemy. Heat is expensive in many data centers. My personal experience has had a Seagate drive running 5C-10C hotter than an equivalent drive from a different manufacturer. When you have a data center full of them, that can be a lot of money.

A $15-20 CAPEX increase per drive with a lower operating temperature and longer lifespan means you are going to save more than that in OPEX over the life of the deployment.I’m guessing Backblaze spends on average $80 per drive, so for that price $15-20 is expensive. On top of that, I don’t know how Backblaze gets datacenter space. Some places charge for power, some don’t. I’m guessing Backblaze is in a data center that charges a flat price per rack, everything included. If you see a flat rate rack as a way to pack as much stuff inside your given power envelope, you’ll quickly find yourself out of power/cooling (same thing) in that data center. Now you need a new data center. Like everything, it’s a balancing act. I don’t have all of the variables in that equation for Backblaze, but the choices they have made are working for them.

This gets to my other real problem with this report: the information Backblaze provides is only of use to people in Backblaze’s requirements space. It’s not generic information, and I dislike how many places hold up their report as the definitive way to buy drives. For almost every other deployment I have been involved with, spending a little more up front for a measurably better product (more reliable, cooler) tends to win in the longer term. This is not true for Backblaze, and that’s fine.

So in summary: You should not buy the cheapest drives possible for use in data center. Consumer drives are targeted at light use consumers for a reason. Don’t hold Backblaze as the barometer for your drive procurement research when you have different operational requirements.

comments powered by Disqus