Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

I don't buy remote hosted storage, so when I saw $12/month for 1TB I was like "whoa, that's way too cheap to be reliable". Then I looked up other services, and some are as low as $4-$5/month for "unlimited" data, with some others being significantly more expensive.

It seems to me these providers are basically a web hoster that specializes in storage, which means 24/7 support, very high uptime, capital investment in infrastructure and ongoing maintenance and planned upgrades, and technical specialization for writing and supporting custom software. Many web hosters get away with low prices by not guaranteeing data integrity and putting a premium on large amounts of storage (or simply over-selling capacity). It's very hard for me to imagine how a company can have a large number of customers, each with 1TB of highly redundant data with high service uptime and immediate support, for $12/month.

For just photos, I would buy the second or third to cheapest option and be done with it, but it still gives me the willies that these prices are so low. The prices of different providers seems almost spastic, running from $1.50 to $150 for a terabyte of backup. Unless there's very good reasons for the variance in price, something seems fishy.



> I don't buy remote hosted storage, so when I saw $12/month for 1TB I was like "whoa, that's way too cheap to be reliable".

The standard market rate is actually closer to $10/TB and that's from the big providers like Google.

> It's very hard for me to imagine how a company can have a large number of customers, each with 1TB of highly redundant data with high service uptime and immediate support, for $12/month.

It's actually not as hard as you think. I built and colocate my own 56TB rackmount and over 4 or so years (the warranty on the drives) it works out to around $2.5 per raw TB, inclusive of all hardware and bandwidth. Optimize that for storage (a lot of my server costs are compute), scale it up, assume the server will live 8 years or so until you replace it and you should be able to get that below a dollar per raw TB. Replicate a few times and you're done.

> The prices of different providers seems almost spastic, running from $1.50 to $150 for a terabyte of backup. Unless there's very good reasons for the variance in price, something seems fishy.

The reasons are variations in replication (for example 3 copies vs 2 copies vs no copies vs 10+3 erasure coding), support, location and bandwidth prices I imagine.


> It's very hard for me to imagine how a company can have a large number of customers, each with 1TB of highly redundant data with high service uptime and immediate support, for $12/month.

Some of them (I remember Tarsnap's page about this) outsource the storage itself to AWS.

https://aws.amazon.com/s3/pricing/

$12 per TB per month would give a slight profit margin on the S3 Standard - Infrequent Access class (currently, a profit of about $2.20 per TB per month) and a significant profit margin on Glacier (about $10.90), with an appreciable loss (about $6.50) if the average customer block exceeded the Infrequent Access restrictions. It might be possible to make the client optimize accesses in some way that makes it less likely that many blocks will exceed Infrequent Access rules, and maybe even to try to keep the typical block in Glacier instead?

https://aws.amazon.com/s3/faqs/#sia_anchor

It seems like an interesting challenge. (Yes, support and administrative costs need to come out of that profit margin.)

Also, Amazon apparently offers "lifecycle policies" to try to do this automatically instead of explicitly. It seems like those policies try to migrate blocks into a cheaper tier that assumes less frequent access if the blocks have not, in fact, been accessed on a certain schedule. A cloud storage vendor using AWS as its backend could then try to optimize manually at the per-user level, or just let AWS do it itself empirically, which wouldn't save quite as much money as correct guesses about what particular users will do with particular data, but would require minimal engineering effort on the vendor's part.

Edit: It seems like it will be hard to compete with the AWS/S3 backend for this kind of service! Now I wonder if there are providers who are known not to use S3.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: