Same. I was even thinking: Genius! I almost never need to restore my backups. If I can backup cheaper but still with the peace of mind of knowing that if I do need to restore I just pay an extra fee.

I wonder if there's a real business model there. Most of us restore an extremely small percentage of what we backup. Could you offer a cheaper storage cost than other businesses and make up the difference on the small number of restores that occur?

There's definitely a business model, but it's short lived: you have to skip the country before the first person tries to restore his data and realizes you were throwing it all away to begin with. :)

Not really, all the cost is in storage and redundancy, the process of reading/writing/restoration is relatively negligible unless people are using it for media hosting like Amazon S3.

If you're really that desperate for cheap backups, you'd actually be best off storing terabyte hard drives / SAN boxes around the country, maybe in colo'd buildings.

I'm not at all sure that you're right about that. The dominant cost of keeping data on-line at the moment is the cost of the electricity to run the server and the air conditioning. If the number of drives in active use is a small proportion of the total number of drives then you might be able to achieve real savings by only having active drives actually on-line. There would be a lot of tricky issues (Where and how do you store the inactive drives? How do you swap them back in when needed? How do you insure data integrity for off-line drives?) but if you could solve those I think there might actually be a business there.

I also think you would save costs on customer support. You would have less features after all. Once a drive / tape / whatever is full you take it offline or power it down.

If you can get the entire operation costing less, then it's fine if restores take longer and cost more... because you pass that fee on to the customer. I for one would be happy to pay less for writes and more for reads, since I read backups far less than I write them.

And it's not about being desperate for savings. It's about reducing the overall cost of a commodity... storage.

well, all you are saving vs. disks that are always active is what, 10 or 15 watts per disk? (we are talking low-power disks, I imagine) I mean, a watt-month isn't free, but even at my scale, you are looking at about $0.19 per watt for just power, $0.27 per watt month for power and rack space inclusive. I'm still small enough that I rent racks from other people, so that number is for the watt of power my equipment uses /and/ the watts the cooling system needs to burn to get rid of the waste heat from that watt.)

I wonder if simply using the backblaze type setup with 'green' drives that are set to 'power down' when not in use would be the best solution? In that case, you'd want to minimize writes as much as reads, but it would make some sense to offer a discount for data that 'just sits there'

On the other hand, I bet that when you buy power 'at scale' you can get your watt-month prices quite a bit below $0.27 (note, you must roll in the cost of both the watt your equipment uses /and/ the watts the a/c system uses to deal with the watt of heat. At the low end, the provider bundles those two, but if you build your own data centre, you need to be careful to build that into your cost assumptions.)

The other problem with doing this at the low end is that where I am, I pay full price for a circuit, no matter how much power I pull off it (the circuit just blows if I get too carried away) and even places that do rent you metered power usually have high costs for rackspace that are not metered, so really, you are only potentially saving two or three bucks a month per disk, even if they were idle all month.

Now, also, 'green' disks are usually really slow. 5900rpm rather than 7200rpm. But they are also really cheap, so if your customers want to accept slow, you can save quite a lot on your CapEx. Like probably half, if you use consumer-grade 'green' drives rather than the 'enterprise sata' I use. Using consumer grade drives has a bunch of other problems that your software will need to account for, too... but that's a place where you potentially could have pretty big savings.

>The dominant cost of keeping data on-line at the moment is the cost of the electricity to run the server and the air conditioning.

That fell under "storage and redundancy".

Storage = the hardware and electricity required to store/minimally serve the data

redundancy = number of places the data is stored

thusly

cost = storage cost * redundancy

I really hate having to verbosely explain things like this. Is it even remotely possible for you to give me the benefit of the doubt and not assume I'm a blithering moron when I don't expound ad nauseam every little detail of what I'm describing?