This gives Amazon the opportunity to spread around the disk across many CPU’s (”Google File System” style) most likely utilizing commodity hardware (read much cheaper as far as hardware goes). Now in order to do this economically it requires a lot of man power (maintenance), which is possible only on large scale.

I personally do believe “@Not Apples to Apples”. You can get a price of storage alone < $0.15/GB on a fairly small scale. Not sure you fit in that category. Can you site the total amount of storage you have?

Now I doubt this can be done using hardware from the big boys (EMC, etc). I would love to see numbers that say otherwise.

As you stated your pre-existing setup was close to $6/GB, but that included RAID. Discounting for RAID 5 (losing 20%) this means about $4.80/GB. The price of Isilon is around $1.5/GB, the price of Coraid is about $0.71/GB (they claim $0.64) - http://www.coraid.com

So this is 3 to 6 times lower.

BTW: The Isilon solution is very scalable and nice, but the main advantage is you get a big NAS device within one file system, which I don’t think your application cares about.

All the prices above however are purchase price for hardware. To compare apples to apples with S3 we have to provide RAID and double the number in order to provide a second copy. Then we have to lease the equipment and account for the cost of financing. This still does not account for the price of replication (at a minimum the cost for bandwidth). Let’s do the calculations for a typical 36-month lease:

If the purchase cost of storage hardware per RAW GB is $1, the true cost will be $1 / 0.8 * 1.10^3 = $1.66375 for 3 years = $0.0462 per month
This assumes RAID 5 (N+1) with 5 disks per array (the 0.8) and 10% per year interest on the financing

So the magic constant for converting purchase cost to monthly cost under these conditions is 0.0462

Now we need to multiply by 2 for redundancy. So for $0.71/GB purchase price we get about $0.0657/GB
You are left with 8.5 cents per GB to pay for everything else, which should be doable. The everything else here includes rack space, power, warranty and repair cost and labor for support.

None of this accounts for other infrastructure. 1 Gbps switches are quite expensive. Not to talk about 10 Gbps.

The way I came up with $0.71/GB for the Coraid box was to spend $4000 for SR1521 Appliance + $4000 for 16 x Seagate Barracuda ES.2 ST3750330NS 750GB at $250 a piece. This leaves you with 1 spare HDD. Total is $8,000 for 11.25 TB.

Notice the cost of hard drives equals the cost of CPU. Now consider if you used commodity hardware. You can assemble a similar configuration for about $2,000 and now the CPU to storage cost becomes 1:2

This obviously is not the same as the Coraid box. It requires additional labor to assemble, test, create OS image, etc.

You could even get BUFFALO 4TB TeraStation Pro II. Not very scalable, but purchase price is $0.50/GB. Granted the RAID loss here would be 25% (only 4 drives).

Now coming back to the initial issue. You are not in the business of managing storage (and other infrastructure), so IMO you are better off with S3. The only thing that has to be considered carefully here is bandwidth cost. $0.20/GB is very expensive. If you start using S3 as the space where you provide data to end users they could sink you. End users in your case are individuals who typically have unlimited download, so they won’t care if they download the same file 15 times. It would be different than using it for backup.

You probably already have good statistics on the amount of bandwidth consumed by end users, so take a careful look there.

Prove it. I have a lot of money to invest in storage, and if you can really deliver a solid storage solution with $0.15/GB/month TCO, including a minimum of 3 geographically replicated copies (so $0.15/3GB/mo), and which will continue to drop in price without leaving me stuck with sunk costs, I’m all ears and the contract is yours. Better yet, prove that you could have done this 18 months ago (I expect an Amazon price drop “soonish” since disks continue to get cheaper).

I’ve looked at nearly every provider there is, including Isilon, and they’re all much much more expensive once you factor in replicated geographically disperse copies, including bandwidth, power, and man hours.

But I’d be thrilled to be proven wrong - and you’d win a multi-million dollar contract. Make sure your pitch is truly apples to apples, of course.

So put up or shut up

Amazon S3 can make sense for small or mid-size companies, I don’t disagree with that. In many cases, the people resources don’t exist within the company to support running even a small data center. But your argument was based solely on hardware implementation costs, and your comparison is not a good one.

At the end of the day, one rule still applies: You get what you pay for. Not only can I implement a solution cheaper than Amazon with those “over-priced storage vendors”, I can implement one that doesn’t go down 4 times in 2 years - http://blogs.smugmug.com/don/2007/01/30/amazon-s3-outages-slowdowns-and-problems/