Werner Vogels, Amazon's chief technology officer, said in a blog post this week that Glacier's price of $0.01 per gigabyte per month made it "extremely cost-effective", and also explicitly described Glacier as an alternative to "cold data storage systems such as tape robots and tape libraries".
But is it that much of an alternative? To find out, I asked tape specialist Quantum to give me a breakdown of the cost of storing 10PB in tape systems over five years.
The total cost of storing 10PB over five years with the company would work out at $669,663.70 (€533,341.61) in total — or $133,932.70 (€106,772.53) per year, Quantum told me.
The majority of this expenditure comes from buying a high-end i6k tape storage array ($450,610.12 or €358,966.08), and the next highest expenditure ($135,198.32 or €107,702.00) is buying the LTO-5 tapes to store data on it. The third is the $75,433.84 (€60,092.28) cost of a support package that provides next business day onsite support in case of failures. There's also a small cost ($8,261.44 or €6,581.25) for installing the equipment onsite.
So how does this compare with Glacier? Admirably. To store the same amount of storage in Glacier for five years costs $6.3m when storing it in the cheapest datacentre region of North Virginia.
I got to this figure by converting 10 petabytes into gigabytes — 10,485,760 gigabytes — and multiplying the figure by the cost of storing a gigabyte ($0.01) to get a cost of $104,857.60 a month. I then multiplied this by 60.
$6m versus $600,000
So, a simple comparison shows us that Glacier is almost 10 times as expensive as an on-premise tape system with support.
However, it's important to note that Amazon has a few advantages, specifically around the rate at which Glacier can ingest data and the fact that all administration is outsourced to Amazon. Those differences would be worth going into further details about if the price difference was smaller, but at a 10x difference, it seems tape is still king for very large storage deployments. The cloud has a way to go yet.
Additionally, businesses pay no upfront costs for Glacier and its fees are levied monthly, compared to a hefty upfront capital expenditure for those going down the on-premise route. This is likely to be an issue for small firms, but such companies are unlikely to want to archive a whopping 10 petabytes (unless they're Instagram, at least).
Readers, what do you think? Does Amazon have a chance of replacing tape, or is the price difference, combined with cloud-fear, just too much? Update: Friday 8pm BST In light of this piece, Amazon crunched the costs of Glacier compared to an on-premises tape solution. Their analysis finds that tape works out to be significantly more expensive than Glacier over the five year period, when you take into account overall storage utilisation, electricity costs, bandwidth and other factors. I understand they will be publishing this information in a blog post this weekend, and once they do I'll come back here with an analysis of the implications of their findings.