Ask YC: 40Tb in a year. Would you use Amazon?
Hi there. We're building a system which stores voice audio files. We're looking at good compression codecs for it (speex) but we're looking at 30 million minutes of audio a month which is looking like 40Tb of data storage a year. These kind of figures scare me a little!! Wondering if you would use Amazon for this or go another route? Just interested to hear from anyone who's doing anything of a similar scale
20 comments
[ 2.6 ms ] story [ 56.2 ms ] threadSure you may not agree with his idea and service suggestions but finding a creative solution to this problem wouldn't hurt. If I remember correctly, the hot or not guy did this (mentioned in Jessica's book). I don't have the book with me so feel free to correct me if I'm wrong.
Look into other CDNs like Akamai or Limelight. I think the bulk price deal you're going to get with an established CDN is better than the rates you'll get with Amazon flat rates.
http://gigaom.com/2008/06/25/structure-08-werner-vogels-amaz...
So yeah I'd probably use them.
http://calculator.s3.amazonaws.com/calc5.html
A quick check tells me that ~40,000GB storage is about $6k/month, not including xfer costs.
There are plenty of options with cloud storage.
I hope you have a way to balance that cash flow.
I don't know enough about their model and what services they provide though. I'm thinking 40TB a year storage, but ~100TB transfer per month or more - which isn't a little.
But my biggest point is future expansion. 40TB in year one... how many in year 2?
All hypothetical, though - personally I think I would go for something like S3.
40Tb can be handled pretty well by S3 and other storage services and they have pretty good pricing information to model your costs. Note that they don't (yet) provide very specific SLA's for data availability, so keep that in mind when designing your system.
Maintaining your own drives with some sort of redundancy (RAID, automatic copies, etc.) or using something like (bias alert) our open-source project http://allmydata.org which is effectively a software RAID layer both require some IT and systems energy, so this has to be bundled into your operational costs if you choose that route.
Just to emphasize what others have mentioned, it is important to incorporate the new data influx rate into your model. If you are successful, 40Tb this year might turn to 120Tb next year, so make sure that your cashflow model can support the underlying cost of whatever system you choose.