“Just let me geek out for a second”
Absolutely IBM storage person, the more geeking out the better!
If we want to do something about it we have to revolt in the streets.
With a pile of random access devices, the time and cost of a single record retrieve is low, though the cost of bulk records can get high.
All the oppressors need is a half decade window, maybe a decade to be pretty much golden, as far as totalitarianism goes.
The tape will definitely hold the data, and then some apparently.
But access costs go up.
The drives seem to encourage indiscriminate targeting.
Tapes would at least favor some priority.
The smarter bears will use both!
You are not wrong about what it will take.
Unfortunately, currently there is no other commercially available method of archival storage, except magnetic tapes. Optical storage has a too low density to be able to compete with magnetic tapes.
Imagine saving BILLIONS of [data points](face,interactions,vids,text,etc) per person on hot or warm? NOPE
Life history will be on tape in the archives of such...
For example, storing your position every 20 seconds might take 10KB a day. You'll collect 15 million data points in a decade, but each one is only a few bytes.
so you subject-A at frame 0 then you have suject-A at frame 100
but no data from subject-A between - but you have data from subjects-b-------------j and can track subject-A in such....
extrap, and you have the "zoom and enhance" of surv....
Edit: but i like your discourse
> The Utah Data Center (UDC), also known as the Intelligence Community Comprehensive National Cybersecurity Initiative Data Center, is a data storage facility for the United States Intelligence Community that is designed to store data estimated to be on the order of exabytes or larger. Its purpose is to support the Comprehensive National Cybersecurity Initiative (CNCI), though its precise mission is classified. The National Security Agency (NSA) leads operations at the facility as the executive agent for the Director of National Intelligence.
Crazy good tech accomplishment right there.
In a way, I am somehow comforted. Tape remaining relevant kind of dropped off my radar. Good to see.
317Gb / square inch AND 15km/h transport speed? That is nuts fast.
Wonder how robust it really is? Of course at those densities and potential transfer rates there is ample room for error recovery.
Not quite sure where IBM's numbers here come from, their previous numbers don't match up to the progression of the LTO tape series' capacity. Maybe they are citing "research numbers" that they can do in a lab but aren't production-ready yet. I would certainly assume they are citing "compressed" data figured there.
But certainly tape has continued to progress much faster than most people would have imagined. Big tape libraries are still a thing in certain environments and they work very well, there is no better solution for bulk cold storage.
LTO is cool and all but is the "compressed capacity" number really something to repeat with a straight face? The tape holds 18TB, we don't need to pretend it's anything else.
> But certainly tape has continued to progress much faster than most people would have imagined.
Mostly it has. But I'm somewhat worried about the future after the sudden late-game announcement that LTO-9 would have a 50% capacity improvement instead of the usual doubling.
No, but it's been the standard in the tape industry for decades. Probably dates back to the first tape controllers that had builtin compression (so compression didn't tax the main cpu).
IRIX allowed for live file system backups and the drive doing compression meant all that happening with negligible user performance impact.
Was literally set it and forget it, aside from tape rotation into off site storage.
Compression would have had an impact.
We don't do multi user app serving much today, so maybe a smart drive has less benefit. But it mattered then. 2000's era.
The hardware is not going to read 300 Gb/in density at bicycle speeds. :)
Me wondering about those micro SD cards that can store 1Tb.
Ad absurdum, stacking two Micro SD cards on top of one another hasn't just doubled your density even though surface area is unchanged.
In theory, write-once memory cards, using some kind of antifuses, could be designed to have a lifetime good enough for archival storage, but nobody has attempted to develop such a technology, because it is not clear if there would be a market for them.
Most people do not think far ahead in the future, so they do not care much about archival data storage, until it is too late and the information had already been lost.
I disagree that it's unusable. You'd end up with a puck the size of a data tape that can archive a petabyte of data and needs to be plugged in to a 5 watt power supply for long term storage. That's not super onerous. Then consider that tapes need to be stored at exactly room temperature with 20-50 percent humidity, while this puck would barely care about environment at all. And you could plug it directly into a computer without a $5k drive. Honestly it sounds pretty good to me. We just need to drop the price of flash by a factor of 20 to make the scenario happen.
The tape stores 580TB for a length of 1255m, does that mean the read speed is 580TB/1255m*15km/h = 1.9TB/s ? Seems too high.
The drive has 32 heads, and reads/writes 32 tracks at a time. It goes from the start of the tape to the end, then aligns with the next 32 tracks and reverses direction.
Each group of 32 tracks is called a wrap. There are 208 wraps, so 6656 total data tracks. Even wraps go one direction, and odd wraps go the other direction.
That's the important part.
But also the tape is divided into 4 "bands", each one holding a quarter of the wraps/tracks. Between the bands, and at the edges of the tape, are special servo tracks that are used for alignment.
So when a source talks about "wraps per band", it's pointless abstraction. Unless you're really in the weeds, the only thing you want to know is the total number of wraps.
It did say that. I wonder if there's some multi-track/double-layer/double-sided stuff going on that needs multiple passes to read, that does sound awful fast!
You could hypothetically do that, for sure, but the software basically is derived from the tape era where you'd just have one logical stream coming out of the tape.
Also, I suppose they limit the read bandwidth with something like what a single Infiniband connection would support; few disk arrays support much higher speeds.
The numbers they cite for "previous generations" don't match up to the progression of the LTO tape series' capacity. Maybe they are citing "research numbers" that they can do in a lab but aren't production-ready yet. I would certainly assume they are citing "compressed" data figured there.
Also bear in mind that tapes typically store data striped across the tape in multiple tracks and multiple bands. There are four bands per tape and 12-52 wraps per band, so reading the whole tape requires up to 208 passes across the tape.
https://en.wikipedia.org/wiki/Linear_Tape-Open#Physical_stru...
But yes, to agree with another parallel comment, tape data rates are quite high sequentially (abysmal in random of course, but that's not how they're used). LTO-8 does 750 megabytes per second compressed / 360 megabytes per second raw.
The table from IBM[1] in the middle of the page says this number is in gigabits (not gigabytes) per square inch. This is obviously impressive nonetheless, but I wonder what else this article got wrong.
[1] https://petapixel.com/assets/uploads/2020/12/ibmtale.jpg
Yawn on the raw density figure, though.
The chip inside a 128 gigabyte micro SD card is a small fraction of a square inch and is cunning enough to provide random access.
You just can't easily and cheaply have a long tape of them.
It used to be the case once upon a time that mass storage media like magnetic tapes and discs based on writing on surfaces with a head had better density than memory chips.
That seem perfectly reasonable. You can get 256 USB flash in in size about half square inch, and that is random access data. (e.g: https://www.walmart.com/ip/SAMSUNG-256GB-Bar-Plus-Titan-Gray... )
> Data on the tape is stored at a record-breaking density of 317 gigabytes per square inch...
> When tape is being read it is streamed over the head at a speed of about 15 km/h and with our new servo technologies we are still able to position the tape head with an accuracy that is about 1.5 times the width of a DNA molecule.