Thoughts on Glacier pricing | Hacker News Reader

Thoughts on Glacier pricing

daemonology.net

51 pointscperciva10 comments

My favorite part of this whole writeup was:

"Epilogue: They each blame the other for the increased bill they're receiving from Amazon, leading to a breakdown in their relationship, and they get divorced a few months later. It was inevitable anyway: Cryptographers never really trust anyone."

Other than that, I think this whole post is spot on. The product pricing is clearly not in line with everything else they do.

I had fun writing that paragraph. :-)

I got so confused by this retrieval pricing scheme I decided to make a spreadsheet to calculate the bill for various usage patterns, feel free to fork it and edit the numbers/scenarios:

https://docs.google.com/spreadsheet/ccc?key=0AnYE99fIo31idFd...

In short, the scheme very heavily penalises doing a large restore from backup, but it's quite cheap if you just spread your restores gracefully across the month.. which makes a lot of sense if they're using tape drives.

I find that an easier way to think about the the retrieval pricing is "bandwidth overage times peakiness", where "bandwidth overage" is the amount of data retrieved in the day minus [5% of data stored]/[length of month], and "peakiness" is the proportion of the day's data which was retrieved in the peak hour (somewhere between 1/24 and 1).

Right.. and it's that peakiness factor that really kills you.

PS: If you were 'anonymous user 21' looking at the spreadsheet around 20 mins ago, sorry I was screwing with things, making all the bills look like $0, it's fixed again now.

The hidden strictnesses (triggering per-day costs even when you are below what is described as a per-month allowance) make me think of "glacier pricing" where glacier is a common noun instead of a proper noun. I.e. 90% of the costs being hidden below the water.

Those are icebergs. Glaciers are inland ;)

Is this actually a case for tarsnap support? If you aggregate lots of people's data but only some of them access it, it's cheaper.

The common use I see for glacier is store something, but retrieve all of it only once per year. Everybody storing their own data pays in that case. But if you get 1/(5%) == 20 people to pool equal amounts, every one of them gets a free retrieval every month.

Yes, in the common case there's an arbitrage potential here -- but as the story of Alice and Bob illustrates, it's possible for the arbitrage to backfire.

User A stores 1 TB and can get out 1.7 GB for free per day. User B stores 100 GB and can get out .17 GB for free per day.

It doesn't make sense to me why this free read amount should be higher for User A than it is for User B, since Amazon is such a "pay for what you use" service. If they both read the exact same amount (read 1 GB), User A gets it for free and User B pays.

I understand that this is paid for from your monthly fee from how much you're storing, but then (again) this isn't "pay for what you use".

Reply on news.ycombinator.com