Tape Storage Might Be Computing’s Climate Savior | Hacker News Reader

Tape Storage Might Be Computing’s Climate Savior

spectrum.ieee.org

40 pointsrbanffy53 comments

This is going to sound pretty damn weird, but the last couple of places where I though tape backups would have been the right plan, we had people who saw the tape industry as a bit untrustworthy. Not the tapes, but the industry.

It took me a while to find out why, but the whole tape storage size not being their actual size has left a long shadow.

It is amazing that something so small can leave such a large bad taste in people's mouths for so long.

Unpredictable compression ratios weren't as much of a concern for me as physical failure. The bad taste was from having to unspool an entire backup tape onto the floor with a hand drill to get the tape reader operable again. All of this just so I can load another tape from the previous week to attempt a moonshot at restoring services before everyone wakes up.

You can test your restores all you want, but in an emergency, there is a chance it's going to jam up like a printer and destroy your backup at the same time.

If you think one tape backup for important stuff is enough, then you did it wrong. Backups for your backups for your backups, is the way.

> Unpredictable compression ratios weren't as much of a concern for me as physical failure.

totally. I just think that their quirk in how they advertise has pushed people away from using them at all is interesting.

As one guy said, "I can compress stuff on my disks as well, it doesn't mean that they have twice as much capacity, and shouldn't be advertised as such"

I thought people would have just kinda shrugged their shoulders and just went "I guess it is just a little weird" but it really has pushed people away.

Wound a 9600 bpi on backwards (twisted tape) by mistake and was made to fix it. That was an interesting day.

I think a lot of people have also been burned by the continuous new rollout of tape formats.

People invest in tape because it offers long term offsite storage, but vendors only offer drive life spans in the order of 4-6 years. After this they rollout higher capacity drives/cartridges and force everyone to refresh their entire hardware and tape collection because they end of life equipment. This costs millions of dollars for a medium sized enterprise.

LTO5 was released in 2010. It offers all the feature of a LTO9 tape (encryption, compression, RFID chips, partitioning, LTFS support) only with much less space per cartridge. These LTO5 cartridges are still readable with very low error rates regardless of how they were stored 13 years later, but sourcing drives is getting more difficult every year.

Tape media needs the equivalent of an LTS Operating System release, where a vendor will guarantee the drives, libraries and cartridge formats will be manufactured, supported and maintained for 20+ years.

I guess most regulations require storage of data for 5 years, at least in banking that's the common case. So you can literally throw with added step of destruction at the end of supported life.

Lifetime of say 9 guaranteed supported years is useless in such model, no added value over 5. If 10 year ones cost 2x the 5-year ones they are still worse deal for company (with added cycle of safe erase to reuse).

Also, "very low error rates" sounds pretty bad - if I need to get back some client documents 95% success rate is abysmal and unacceptable, even 99% is NOK for most use cases. Unless you mean some self-correcting ones, but then we still have 100% reliability.

if I need to get back some client documents 95% success rate is abysmal and unacceptable, even 99% is NOK for most use cases

Parchives [1] solved this problem a long time ago.

[1] https://en.wikipedia.org/wiki/Parchive

On the other hand, the weak commitment to backward compatibility enables new drive/tape technology to move forward. It’s a good thing that we have LTO9 densities.

LTO drives have two generations of backward read compatibility, and LTO7 drives are readily available, so those 13-year-old LTO5 tapes aren’t stranded yet.

13+ years is pretty good when most enterprise equipment gets depreciated after ~7 years.

LTO5 drives are readily available on eBay; units designed for tape libraries frequently sell for $100 or less and are easy to convert for external use.

Specifically, I've converted library drives from IBM and Quantum libraries, both with IBM mechanisms. Drives from IBM libraries are a bit easier to convert, but Quantum units have a smaller footprint.

In either case, once you disconnect the proprietary library power/control PCB, the drive and chassis fan can be connected to a standard PC power supply using off-the-shelf molex cables; I/O is standard Fibre Channel.

For IBM library drives, this is all that's necessary; the Quantum library drives I've converted require a firmware setting change to enable the FC interface on power up:

https://github.com/AC7RNsphnHVbyT4/ibm-tape-drive-automatic-...

TL;DR:

(1) Connect RS-422 USB adapter to a computer and the drive (N.B.: Wire colors from the article are not standardized; you can easily find the correct pinout by tracing the signals from the serial transceiver IC on the library interface PCB to the drive interface cable).

(2) Power up the drive.

(3) Send the byte string from the bottom of the drive until the drive reboots.

At this point, the FC interface should activate on power up as with any other standalone drive.

Finally, IBM supplies a useful diagnostic tool to ensure tapes and drives are in good working order (free IBM account required for download):

https://www.ibm.com/support/pages/ibm-tape-diagnostic-tool-i...

All of what you have said is correct.

None of what you have said is going to happen at a bank, insurance company or media shop who need to comply with all the corporate legislation post-Enron.

> People invest in tape because it offers long term offsite storage, but vendors only offer drive life spans in the order of 4-6 years.

The exact same thing can be said of other storage technologies (HDD, SSD, etc).

Do you know why they do it? Because more bits keep getting created and no one is throwing away the previous bits, so they just keeping stacking up on each other.

SATA was released in 2000. You can still buy new drives or motherboards with connectors. NTFS was released in 1993. You can still use it with off the shelf operating systems.

If I have an NTFS formatted SATA drive from 2002 I can plug it into a brand new machine and it will work out of the box (except it won't because it would have been dropped)

LTO6 was released in 2013 and is end of life. I cannot buy a new drive to read it. Storagetek T10000D was released in 2013. It is near impossible to find drives to read them. IBM TS1140 was released in 2014. Again all the tape hardware is EOL.

My point was: why is it that a media designed for long term storage, doesn't have long term support. If you want more uptake of tape, support the hardware for as long as you claim the tapes last.

> but vendors only offer drive life spans in the order of 4-6 years.

You should read and transcribe tapes from time to time anyway. I would be scared if I saw a vital tape that hasn't been read in 13 years.

Tape companies are lying scum.

Computing's Climate Savior - really? By their estimate, if every organization in the world put in the effort to transition 60% of their data to tapes, you'd save 79 million tons a year, a whopping 0.16% reduction in CO2e. This doesn't even cover the current rate of emissions growth or the steady increases in hot data volume. No mention of Jevon's paradox either - highly likely that companies will take those IT savings and roll them into more IT resources.

It's still worth considering tape storage. But if you think you're "saving the climate" by doing so, you should tone down the greenwashing - don't want to pull a muscle patting yourself on the back.

This is a small amount, but it is not negligible.

If we really wanted to tackle green house gas emissions for climate change we should ban cruise ships, as they are equivalent to 1 million cars each [1]. Then heavily investigate and possibly prosecute large companies in a plethora of sectors.

[1] https://www.cbc.ca/radio/asithappens/as-it-happens-wednesday...

> ban cruise ships

Or force them to use carbon-neutral fuels. Or get serious about nuclear-powered civilian ships - the NS Savannah was a good technically successful example.

And, while we are at it, remember a nuclear (or even carbon-neutral) airliner is a challenge, but a nuclear transatlantic ship is not. Add a hydrofoil and we'll be in the Thunderbirds future.

Doesn't touch on air conditioning, the fujifilm calculator lists cloud storage as vastly cheaper than disk-based storage, which AFAIK many would disagree with.

> HDDs have a lifespan of approximately five years

For an active datacenter, no?

And speaking of active, active storage lends itself to monitoring and automated healing a lot better than tape archives.

What does automated healing mean in this context?

Assuming a sane RAID situation, e.g. data being duplicated onto multiple drives. If one of those drives fails, making additional duplicates onto the surviving drives, or rebuilding the data of the failed drive onto a new drive.

Can also be done with tapes, but you would require multiple tape drives (though if you can afford an automated library, thats probably nothing in comparison), additionally you can't find the failures until you take them out of "cold" storage.

If you can respond to the first failure when it happens, you can react before being hit by later failures hitting the duplicates. Downside is, since you have active components, failures are expected to be more common.

HDDs designed for industry have a lifespan of longer than 5 years in typical data centers. In fact, recently there have been many teams across industry expanding the lifetime of HDDs past 5 years well into 6 and 7. You can also check out Backblaze's drive stats to see how many drives are actually going past this time span: https://www.backblaze.com/blog/backblaze-drive-stats-for-q1-...

And one would expect actual data center grade HDD's to do even better.

Backblaze is fameous for using piles upon piles of consumer grade discs in their offering. Which is why their stats are so valuable to folks who are not running data centers, but might want a reliable drive for their NAS.

Tape libraries all have active monitoring and have for a while.

But really, it's mostly about scale. Tape media is dense in a way that spinning disk can't catch up to.

~350cm³ for a tape, ~375cm³ for a 3.5in HDD, With HDDs being able to extend past the 18TB of LTO-9 tapes.

I am too lazy to measure them right now, but your numbers must be wrong.

Visually, an LTO cartridge cannot have much above a half of the volume of a 3.5" HDD and it must be at least 3 or 4 times lighter.

Carrying a suitcase with twenty LTO cartridges is as easy as carrying any other typical suitcase, but carrying twenty HDDs would not be easy at all, due to the great volume and weight, and the volume in a real situation would be greatly increased by the need of adding some serious padding, in order to avoid catastrophic damage if the suitcase were accidentally dropped.

More importantly, the price per TB of tapes is much lower, and when you consider the ratio between the expected lifetimes, the price per TB and per year is much, much lower.

Of course, tape drives are extremely expensive, so you need to store at least a couple hundred TB to break even with the price of HDDs.

You are right. double checked, I think it might have been some site listing the box size for some unknown reason, 260cm³ should be the correct size.

Your analysis ignores cooling.

LTO-9 can cram 18TB uncompressed data onto a tape. A 3.5" HDD is a bit larger, but we already see 25TB drives. They are much heavier though, so LTO has an edge on density per mass, while HDDs have an edge on density per volume. Tapes (specially in tape library) have a huge advantage in density per dollar as capacity increases linearly with the number of tapes.

OTOH, HDDs have much better latency, as even a powered down hard-disk will power up and retrieve data faster than the time it takes to load a tape and read it. I can imagine a number of ways to partition a HDD pool in ways that the access patterns allow for most drives to be unpowered at any given time.

Having been responsible for several multi-petabyte storage projects, my experience is that TCO over a 20-year archive lifespan (LTO's design spec is 30, but boundary-pushing is not a good idea in the long-term storage game) is much lower for tape. Indeed you can't generally get reliable parts and support for disk shelves beyond about seven years, which means a full-solution refresh at least twice as often as with tape.

Most of the time the right answer is most of the storage in tape, with a couple of petabytes of disk in front of it to deal with hot data. Even in that circumstance you have to carefully manage the disk storage and filesystem or you wind up with client devices too new to access the supported fs version, another problem that basically does not exist with tape.

That's a good point - there is no such thing as a tape-only data storage system - you'll need RAM, flash, and HDDs to help ingest large quantities of data so that tape drives can have data ready to write.

Funny story: glacier was originally built on HDD, but the economics went sideways and some speculation about power saving techniques didn’t pan out. So to stop the bleeding aws moved glacier into s3, but kept a fake api with fake sleep statements and very real losses due to the pricing. During that time they reimplemented glacier to be the tape backed up version we have today. Bet you never even noticed. :-)

Is this confirmed? I even saw some speculations that it runs on Bluray.

I was there :-)

Within Google, there used to be significant usage of tape storage until there was some major data loss incident on HDDs and they had to actually read the tapes and someone did the math and realized it would take between weeks and months to read out all the data that they needed.

i guess it depends on the value of the data whether this tradeoff is worth it.

was the data at google recovered or did they abandon the plan and consider it lost?

It was recovered and it was such a pain that they deprecated tape storage entirely.

Call me crazy, but there must be a way to store data indefinitely, right? I mean, writing it on a piece of paper and putting that paper in a box can last practically forever, given its not actively physically destroyed.

Could we not come up with a tape storage that is write-once? Like a PROM, but in tape form? You do this, you add some checksums every now and then, and you just lock the thing away. How is that not going to last 100+ years if physically undisturbed?

Gold film with holes lasered into it should last indefinitely. Probably the longevity of the reel, container and building are more of a problem.

Github's backup on Svalbard also uses some kind of film, and it's supposed to last 500 to 1000 years: https://en.m.wikipedia.org/wiki/Arctic_World_Archive

We have data from thousands of years ago that has survived on stone tablets, papyrus, etc., but we don’t know what the failure rate is. The surviving items might be 1% of the originals.

The best way to preserve data long term is to make lots of copies, distribute them widely, and migrate the copies onto new technologies periodically.

The Rosetta Project had (has?) the goal of storing a modern day Rosetta Stone[1] on nickle disks etched with nearly 14,000 pages of text in eight human languages. They expect(ed) that with minimal care the discs should remain legible for thousands of years.[2]

[0] https://rosettaproject.org/

[1] https://en.wikipedia.org/wiki/Rosetta_Stone

[2] https://rosettaproject.org/about/

Paper in a box does not last forever--less than 100 years. It can last longer with the right paper and in climate-controlled environment. But this is not passive storage, you have to use energy every day to maintain the facility.

It’s interesting that they don’t compare to flash.

SSD’s also have a much lower carbon footprint than HDD’s. The electricity it takes to keep the hard disk platters spinning dwarfs the footprint of the NAND, including manufacturing.

I don’t think global NAND manufacturing is anywhere near the same storage capacity as disk or tape, yet, but I wonder when those curves are set to cross.

SSD get mixed signals in the reliability/heat stakes. They do fail, they fail in ways we haven't got used to yet, and so TRIM/SMART is only partially ameliorative.

I want to be there. I'm told they run hot, so incur much the same cooling issues HDD do. I am told if you don't recharge them periodically, they are not reliable offline storage disconnected from power for various reasons.

And of course, there's price.

price is coming down fast and coupled with density it is becoming a realistic option

it's already possible to get a 2.5 ssd with 8tb for $350

I'm basically a cheapskate and paid $80 for 4x 2tb HDD and they've hit $120 as reputable SSD and stuck. I keep hoping they'll drop lower but I need SATA form factor and I fear they will only drop in m.2 form now, SATA is dying. SATA to m.2 carrier boards are $20

2.5" sata SSD are falling in price about about the same as m.2 nvme sata, here's a 2TB for $62 [1] (not an endorsement, just an option; always be confirm your ssds have the features you want: bits per cell?, dram?, controller?, secure erase?, etc)

m.2 sata is available at similar prices too, but many fewer choices. mSata seems to have topped out at 1TB

[1] https://www.amazon.com/dp/B07Q37V1C9

US/AU $ works out my price at $92. Plus postage. But, encouraging!

SSD's are not suitable for reliable long term cold storage due to their finite data retention. Newer chips with even more voltage levels per NAND cell for improved density are even worse.

Some of the data center grade SSD's have rated data retention of 3 months at 40°C if unpowered.

Having to keep the archival media powered to retain the data does not sound too attractive.

How much power would it take?

For what I could find, Samsung 970 EVO Plus (1 TB) consumes 40 mW when idle. This means that a 100 TB array would have to spend about 40W on idling, or 350.4 kWh a year.

Even at $0.50 / kWh, it would cost $175, a trivial price to pay for having your data instantly available.

The price per TB is not great though, about $50, while an 45TB LTO9 cartridge is about $110, or $2.44 pet TB.

So it's not the price of the power.

>while an 45TB LTO9 cartridge is about $110, or $2.44 pet TB

45TB is the 'compressed' capacity, the actual capacity is 18TB for LTO9, for $6.11/TB - still an order of magnitude cheaper than SSD, mind you!

I still use discs for archive purposes. Just as it is described by

https://www.canada.ca/en/conservation-institute/services/con...

Reply on news.ycombinator.com