Hacker News Reader

Comment by oliwarner | Hacker News Reader

I was thinking —like other top-level comments here— that it's a great shame that consumer level drives have died out. They're thousands of dollars and all wired for enterprise interfaces.

But I'm really surprised that nobody[1] is offering "tape backup as a service". Pay $30 per 3TB cassette (slightly over market rate), a $10 loading fee and then $1/h for 100MB/s write access. When you're done, pay another $10 and it's shipped to you. Or $5/month for fireproof storage.

This might have to be a pre-booked service, or they could buffer upload onto spinning disks and write to tape after the fact (much faster), but for storing more than 1TB, this method could be vastly cheaper.

[1]: That one quick Google could find.

View on HN

How would this, in effect, be different from any serious backup service in existence today? I'm sure they do their backups of your backups on tape, and at least rsync.net is willing to ship physical media with your data to you.

Price!

In the simplest version of my post, you're paying for media, the time for somebody to pop it in the machine, the hire of that machine while you copy data to it over the network, and the postage of a tiny little LTO back to you for archival.

On higher density tape, you could be looking at a one-off $100 total for writing 10TB of data to tape and mailing it back. There's a lot of room between that price and the nearest incumbent to make a strong profit there.

rsync.net seems very bespoke, so they may be able to give you a tape, but you're looking at $250 per "incident" plus hardware. They waive that fee if you store more than 100TB ($2k/month).

Yeah, they're much more available but most of us aren't after cloud storage, we want a lasting backup in case the NAS dies, or the house burns down. Restore speed doesn't matter.

Ah, true. I didn't consider the "one time dump of a boatload of data" as a use case. My backups tend to run relatively often, and at that point, $10 for each tape load would easily make the tape service more expensive!

AWS Glacier provides a pretty cheap[0] backend for such a service. You could store 1TB for 1000GB * $0.004/(GB * mo) = $4.00/mo, which is dirt cheap, all things considered. Retrieval is about 10x that, per GB, but generally if you are storing backups, you aren't concerned about retrieving it too often, so that cost is acceptable. Compare this to the S3 bucket costs for the same storage size($10/mo for infrequent access, and $24/mo for standard), or to google drive 1TB tier ($10/mo, but no API as far as I know).

[0] https://aws.amazon.com/glacier/pricing/

That's per month... In just 9 months Glacier becomes more expensive than a physical SATA drive(s) of equal capacity. The longer you keep the data, the more it skews against "the cloud".

I'm definitely not claiming to have all the answers here but there are points where tiering your storage strategy to get it onto something stable and without running costs makes sense.

Indeed, at 1TB scale, cloud storage can sound "dirt cheap", and, of course, it is, since the fixed cost of hardware to store that reliably is the same as for 100TB.

10x that, or a petabyte, would cost $144k-$180k after 3 years, depending on region. Hardware [1] might be, generously, $60k. The hardware would also have operating costs of another $26k [2], but Glacier would have retrieval and data transfer costs if one ever wanted the data back.

Of course, tape, the subject of TFA, is even better. $17k [3] for the hardware and minimal operating cost in the labor required to change the tapes.

VC-funded startups have been tending to ignore this artithmetic (which applies to all cloud infrastructure, at only slightly larger scales), lately, so cost isn't always an issue any more.

[1] Let's say 100 12TB drives for 84 data, 12 (triple) parity, 4 warm spares. 3 enclosures, 6 SAS HBAs, plenty of CPU and RAM for ZFS.

[2] Generously, the above is 2.7kW. Adding 50% for cooling, at 12c/kWh. A commercial datacenter could double that (or triple or more, but, for backups, presumably one would find a geographically cheaper one).

[3] LTO-8 drives are $2700 and 12TB cartridge are $170.

> Of course, tape, the subject of TFA, is even better. $17k

> LTO-8 drives are $2700 and 12TB cartridge are $170.

What I initially ignored was the cost of a "library" (robot), which is, arguably, necessary if the minimum time to write a tape is over 9 hours (longer than a standard workday, meaning 1PB could take 17 workweeks to write with a single drive with no robot).

Including a library to hold all 84 cartridges might be tricky, but it appears there are 80-slot ones in the $12k range, which brings the total tape price up to $29k and reduces the write time to under 5 weeks.

Alternatively, one could just buy multiple tape drives and no robot. $25k gets faster write time than above.

It's not cost effective with the rise of the cloud and broadband.

What isn't? Home tape drive ownership, sure. But what I'm proposing (essentially hiring a remote tape drive, getting the result back in the post) is vastly cheaper than cloud storage.

Reply on news.ycombinator.com