By Glenn Fleishman
November 25, 2014 10:28 AM PT
The crashing price of storage
It’s tricky to discount the future, except when it comes to technology. In nearly every way, the march of computing power, memory, hard disk storage, screen quality and the like is toward ever more, ever cheaper. In general, the price of food, shelter, and energy increase over time in absolute terms, while the price of things that contain electronics decrease in real terms. (The big exception is bandwidth in America because of a severe market failure that preserves false scarcity.)
So I’m cagey about getting locked into a price for anything when I know it will, nearly invariably, cost less and be better if I wait. This is why I haven’t upgraded my aging-but-still-reasonably-functional mid-2011 MacBook Air, which has 4 GB of RAM and can’t be expanded to more, because I know a Retina version is coming if I only wait long enough, likely only costing slightly more than today’s Air.
But when I spotted an offer from Code 42, makers of CrashPlan, for a pre-Black Friday sale (now since expired), I leapt at it. This sale for existing subscribers took about half off the price of its unlimited storage family plan, which is usually $150 for one year or $290 for two. The reason I paid says a lot about the current dynamic in the world of computer storage.
I currently have more than 1.5 TB stored on CrashPlan, accumulated over several years, not including deleted files that are retained as well. (I also back up to local drives using CrashPlan, use Dropbox for key documents, and use SuperDuper to clone my office computer nightly.)
While I had 10 months remaining in my family subscription, I immediately paid $145 for a two-year extension. (Some people intensely dislike Code 42’s reliance on a Java-based client or find its configuration funky. I’ve had trouble here and there, but it’s saved my bacon many times, including restoring about 600 GB from a local backup after a drive failure just a few days ago.)
It’s a smart deal for CrashPlan, because they retain me as a customer, get my word-of-mouth promoting it, and receive the cash immediately rather than in 10 months, at which point I would be more likely to look at the $150 renewal and perhaps not renew at all, switching to another option. They’ve captured me, and while they now have a future obligation to me, they know that their costs in the future will be lower—they’ve “paid” me a retention fee that’s less than the opportunity cost.
In the few years that CrashPlan has been in business, they’ve been able to drop their prices enormously, as have other hosted backup services. In 2008, CrashPlan charged $170 per year to back up 200 GB from a single computer. By 2012, this price had dropped to $50 per year for one computer. They later raised this slightly (to $60 per year). During that same period, the cost of hard drives dropped ridiculously: one popular source puts the cost per GB at 60¢ in early 2006; it’s now 3¢, or 1/20th as much.
Amazon’s S3 storage tracks this drop. When S3 launched, the company charged 15¢ per GB stored—it’s now 3¢ per GB per month. However, consumers have to pay for the cost of an entire drive, no matter how much they use. Amazon charges only for the storage that you’re occupying. Thus the cost of Amazon storage is 36¢ per year per stored GB. S3 is also designed to be resilient, with Amazon making redundant copies and claiming “99.999999999% durability” over the course of a year. (Google charges roughly the same for its Cloud storage.)
Storing 1.5 TB of data would cost about $45 a month on S3’s standard storage. They have a “deep freeze” version that’s only a penny a gigabyte, but still $15 a month. (The data transfer inbound is free.)
This would suggest that CrashPlan can operate its data centers more efficiently than Amazon’s. Code 42 doesn’t provide numbers, but its competitor Backblaze says by using standard drives and custom software—just like CrashPlan—its costs are 25 times lower than S3.
All online backup services also benefit from both scarce bandwidth and uneven distribution among users. The former situation means people restrict backup sets, while the latter means that not everybody is storing a ton of data. While I’m occupying 1.5 TB of redundantly stored data, other folks might be storing 100 GB or even less.
CrashPlan also makes more money from me over time. Yes, they have to reserve capacity for me and increase it over time, but my dataset is growing relatively modestly compared to the drop in their cost of hosting it.
We can get an idea of ongoing capital costs by looking at Backblaze again, which publishes reliability data each year of the various high-capacity drives they purchase: they had 34,881 drives in operation as of September. Failure rates among most drives sit in the low single digits per year, with some outliers. But even assuming a 10 percent replacement rate each, it doesn’t cost them much. Right now, a typical 3 TB drive costs about $100. My 1.5 TB of data is stored at least in triplicate, I’m assuming, thus taking up $150 of drive space exclusive of CrashPlan’s overhead.
But each year, it costs CrashPlan only $15 in drive replacement to maintain that. Even if I increase my data set by 30 percent each year, I’m still only adding about $30 in capital, and drive costs continue to drop. I probably push a few hundred GB of updated data each month, which costs CrashPlan a few tens of dollars in bandwidth. (Throughput costs can be extremely cheap. Amazon charges nothing for inbound S3 storage, and 12¢ per GB for outbound transfer. But I can get 2 TB of monthly transfer for $10 per month as part of the least-expensive virtual server hosting package with Linode.)
This all adds up to a reasonably sound model for CrashPlan to profit off me, even with my large dataset. Some folks who shoot a ton of video and have fiber-optic Internet might be adding a terabyte a month, but they would be outliers. For most of CrashPlan’s customers, the company can make a tidy return even by discounting the future. And I can have peace of mind at half the previous cost.
[If you appreciate articles like this one, help us continue doing Six Colors (and get some fun benefits) by becoming a Six Colors subscriber.]