Cheap and very very reliable archival storage (essentially no failure modes other than the magnetic media itself), weight and volume wise pretty much still the best density you can get 18TB (up to 45 with compression) in a small 200g package is hard to beat when you need to move peta bytes around.

Also tape is much cheaper to produce than platters especially at large volumes.

How does such tape compare to 2" and 1/2" analogue reel to reel tape, in terms of production costs?

A new reel of Ampex 2" tape is $375.00 currently: https://vintageking.com/atr-magnetics-studio-master-tape-blu...

I don’t know but over the counter direct purchase single unit LTO 9 tapes are about £70-80 each in the UK for 18TB, in large volumes they are much cheaper, I’ve seen quotes in the £30 per tape and that wasn’t at a cloud provider or even massive enterprise scale or anywhere close to that.

"Never underestimate the bandwidth of a station wagon full of tapes hurtling down the highway."

For big data transfers, this still holds true.

Tape has a tremendous density advantage over disks. Think about how much area a tape occupies if you were to completely unroll it, and compare that to a disk platter. The flip side is access times are extremely long compared to disks, but for the usual application of backup and recovery that doesn't matter much. In this case they're pairing the tape with faster non-volatile storage so writes at least will be very quick (reads not so much, 2 minute access time!)

The 2min number is likely an average over a large volume of drives.

An LTO 9 tape has over a kilometer(!) of tape in it.

You can spool the tape only so fast without risking damage to either the tape itself or the cartridge so if your file is at the end of the tape you’ll be waiting for a while.

It's a single tape. Even without any kind of fast forward, the write speed of LTO-9 is 6 meters per second. It takes three minutes to go completely end to end.

I think makes perfect sense for archiving. I keep hearing tapes last 30 years, while in the same room temperature SSDs last a few years. Just wish I could find a solid source for it!

A worn-out SSD is supposed to be able to preserve its data for one year for consumer SSDs or three months for enterprise SSDs (not at the same temperature). That's the industry-standard threshold of data retention used to determine how "worn out" is considered end-of-life. A SSD that hasn't burned through its entire write endurance will have considerably longer data retention.

We're not talking about write endurance though? From what I've read, unpowered SSDs only hold the data for a year or two, regardless of the age of the drive.

> From what I've read, unpowered SSDs only hold the data for a year or two, regardless of the age of the drive.

You've probably been reading bad sources. There is a very real relationship between the number of P/E cycles that a piece of flash memory has been through and the unpowered data retention it can provide. Lots of publications and even more random forum posts repeat statements about SSD data retention without including the caveat that the numbers apply specifically to worn-out drives. Brand new flash memory can easily have unpowered data retention in excess of a decade.

If you're reading a source that claims SSD unpowered data retention is only a year, and that source does not qualify that with a statement about the prior condition of the drive, you should assume that you're reading another misinterpretation of the JEDEC standards.

You can buy LTO tapes from Amazon and office suppliers easily.

The drives and the software are the problem.

Drives are very expensive and there isn’t much non enterprise software to manage tape backups.

Enterprise software is expensive and you won’t likely be even able to buy it and since cloud storage and NAS became commoditized all the backup software that targeted small businesses pretty much either died off or went cloud backup only.

Backups or satellite imagery is my guess.

There are no downsides to keeping off-cloud physical backup sets of most critical data in a firesafe or two (and still needed for SOC 2 compliance I understand)

> Aug 23: CloudNordic said: “The attackers succeeded in encrypting all servers’ disks, as well as on the primary and secondary backup system, whereby all machines crashed and we lost access to all data.”

In the 1980s or 90s, Jim Gray predicted that in the far future memory hierarchy would be reduced to main memory + tape.

Not sure that prediction holds up (although streaming dbs in the cloud seem awfully "tape-like" to me?), but I did attempt to quickly redo his "5 minute rule" calculations: https://news.ycombinator.com/item?id=42125052

If I did the math right, there's no point to swap or paging anymore, as the break-even point I got was evict after 1 week* (instead of 5 min) for random access patterns, and evict after 5 hours for sequential access patterns.

These days maybe those breakeven calculations would be more useful for deciding how long to keep something in a register and when it'd be better to recalculate or spill to main memory?

* of course, if you keep everything in main memory and forget nothing, we're close to what in 1998 was one of McCarthy's pipe dreams: Elephant2000 http://www-formal.stanford.edu/jmc/elephant/elephant.html ; I guess if I were to attempt something like Elephant2038, it'd wind up looking like a bunch of past-oriented LTL formulae?

(the "speech acts" were probably Winograd's influence?)

Not to needlessly alleviate your suffering, but what is LTL?

Of course, these Nobel Bauble driven days, “quantum memories” sound like little more than search engine bawdlerised AdWords..

Linear Temporal Logic: https://www.cs.colostate.edu/~france/CS614/Slides/Ch5-Summar...

(for Xφ, I prefer to define it as relative to a given frame: φ holds the next time any of the frame variables change, which allows both continuous<->discrete translation and for Lamport's stuttering [to allow abstraction!])

Next up: GC via creation and annihilation operators? (in my algebraic view of informatics, one is often either structuring by cleaving two parts to make a whole, or destructuring by cleaving a whole into two parts)

EDIT: and by past-oriented I mean using "was ever" for "eventually" and in the past for "globally", as well as "since" for "until". multiply t by i^2.

Thanks for the ref!

Not sure when i’ll get there praying for a serendipitous intersection of cargoculting nonsssociativity and diagram spelunking (aided by your hints, half-remember’d)

Although… i confess to doing the trivial thing of drawing the “cayley graph” of {rock,paper,scissors} and observing that it looks like a loopfree 3-vertex term (1 outgoing)

https://en.wikipedia.com/wiki/Commutative_magma

Rock-paper-scissors also gives rise to scenes that are bad for naive sorting (for graphics or CAD), where you have three objects such that any two can be sorted in occlusion order, but not all three.

Can we triage our ikigai such that I (sympathising with my machinery) take all your overly-associative structures and you (communing with the universe?) take all my insufficiently-associative structures?

(there are many in maths: eg topologies, being families of sets, can be represented by a particular magma)

Lagniappe: https://www.georgehart.com/triad.html

Well ring me when you've a way of categorizing some of the nonrepresentable topologies, I do love any chance to juxtapose those ikigais

https://news.ycombinator.com/item?id=42106477

Categorically, while different algebraic structures may not be isomorphic, they're likely all capable (under decoration) of carrying the same information.

For instance, in the Boom Hierarchy* (sets,bags,lists,trees) we can represent the information carried by any of the structures by any of the others (exercise!) despite the fact that operations on that information may be made easier or more difficult by the presence or lack of laws (idempotency,commutativity,associativity) that are the linear delta-edges between those types-as-vertices themselves.

[IIRC, magmas would be mobiles, and I'd bet we can extend the above to them as well.]

* I know I repeat these often, but hey, I'm not alone: gopher-the-language was inspired by a "go" lemma-function in someone's sol'n to samefringe.

How much work is "likely" doing..

I'd been at peace if you mentioned the triples, but it's too late now..

"likely" is not doing much work for the domain I was thinking of: structures which are suitable for making into collection data structures, in that their carrier sets are infinite because their ops don't lose information about the values (the hyle if not the morphe) given by their arguments.

Of course, absent those restrictions, it's doing way more work than it can possibly handle.

wasn't it our own Carl Sagan who suggested looking into \pi for alien messages (but/and he was perhaps under the influence)

But not the references inside apple boxen

Could be, although I could also imagine the BHA was intending that as a smoke test: if we were to find a putative alien message, we'd want to make sure the same procedure doesn't find other messages in the first N digits of \pi or related constants.

(shouldn't all messages be in \pi sooner or [much, much...] later?)

[does the def'n of "message" include finite length?]

I'd imagine, though, the KST-patched protocol would be to look at how the certain later digits of \pi vary "in realtime" based on where certain earlier digits (of \hbar) tell us to look.. that'd indicate sophisticated control of remote curvature :)

>Sagan's friend physicist Kip Thorne gave Sagan ideas on the nature of wormholes when Sagan was developing the outline of the novel.