Ransomware attacks will become next to impossible because the data will be stored in quartz glass by "...permanently changing their physical structure". So this means that the data is written permanently and cannot be modified, and thus, cannot be held for ransom unless the physical hardware is held for ransom.
Anyone familiar with this, can you please explain how much data can be stored per cubic centimeter? Not sure if this is a good measure of data density, but since we are talking about crystal and layer stacking I couldn't help but think of the physical size in 3 dimensions.
I'm a bit confused on how this could scale - if the way it works makes permanent alterations to the glass (which the article mentions and I presume would be the reason it is resistant to ransomware), what does it do when data changes? Or this a write-once, read-multiple-times solution?
Edit: After reading MSFT's official site on this, it is meant more for extremely long term archival of data, and is intended to be Write Once Read Many which makes much more sense now
I think the point here is that other WORM solutions are not indefinite/long-lasting and expensive so companies don’t bother backing things up, and this should incentivize them to do so. No idea about the cost of this though.
If the main selling point is that it's non-rewritable, it seems like this could be accomplished with existing storage tech. For example why not have a spinning rust disk with a physical interlock switch to prevent writing? Then the disk could still be manually wiped and reused. For important textual data, burning DVDs should be totally sufficient.
The big differentiator is long term durability. Quartz glass will be stable and readable for millennia, whereas even DVDs have a relatively short shelf life. I agree with your analysis though, the majority of use cases could just use existing tech with physical write protection.
From the article it seems to me, to be a backup solution only. Kinda a replacement for tape, with the difference that it is faster writing. What got my attention:
> The images are then sent to be processed and decoded, which leans on machine learning model to convert analog signals to digital data.
Looking at the paper [1] (section 3.2.):
> Machine learning models are able to better learn and account for any noise properties inherent in the end-to-end write and read processes including: inter-symbol interference between adjacent voxels in the glass, scattered light from neighbouring layers during readout, variability between optical components, and more. By contrast, traditional signal processing techinques require extensive understanding of all these characteristics and careful (often hand-crafted) processing to remove their effects.
It feels weird to me, that they target "sensitive industries including finance, scientific research and healthcare", but decoding/reading the data is not really deterministic.
That doesn't sound good in terms of future-proofing this technology. If I want to read stuff 200 years from now, the knowledge necessary for that process will be buried deep inside some opaque ML blob.
When they say machine learning, do they really just mean error correcting codes and convolutional encoding? Is this just a case of using buzzwords or is the ML just doing RS/Viterbi?
This is fascinating. Imagine non-rewritable optical storage media… it is like something out of a Foundation novel! Too bad that that concept is so thoroughly untested, we may never know if anything like that could gain wide adoption
I couldn't tell. Had I made this joke myself, I would have dropped the irrelevant Foundation reference and made it more obvious by wondering aloud if this "hypothetical" device might rotate somehow, or be used to store music.
Perhaps such a device would be named a "Concise Disc".
Absolutely -- but I'm talking about signalling your intent to make a joke.
Obviously the less overt signalling you can get away with, the better the joke is for those who get it. But if you provide so little that people confuse the joke for genuine ignorance, the joke isn't working as well as it could. I'm not the official joke arbiter by any means, but I did read your comment and think about whether you intended this specific joke, and concluded that you probably didn't.
Just in case you weren't joking, and at the risk of making a fool of myself if you were and still are joking (and/or trolling us now), where you initially said "non-rewritable optical storage media", some people people including me thought you were making a joke referencing this: https://en.wikipedia.org/wiki/Compact_disc.
26 comments
[ 2.2 ms ] story [ 71.0 ms ] threadAnyone familiar with this, can you please explain how much data can be stored per cubic centimeter? Not sure if this is a good measure of data density, but since we are talking about crystal and layer stacking I couldn't help but think of the physical size in 3 dimensions.
Edit: After reading MSFT's official site on this, it is meant more for extremely long term archival of data, and is intended to be Write Once Read Many which makes much more sense now
> The images are then sent to be processed and decoded, which leans on machine learning model to convert analog signals to digital data.
Looking at the paper [1] (section 3.2.):
> Machine learning models are able to better learn and account for any noise properties inherent in the end-to-end write and read processes including: inter-symbol interference between adjacent voxels in the glass, scattered light from neighbouring layers during readout, variability between optical components, and more. By contrast, traditional signal processing techinques require extensive understanding of all these characteristics and careful (often hand-crafted) processing to remove their effects.
It feels weird to me, that they target "sensitive industries including finance, scientific research and healthcare", but decoding/reading the data is not really deterministic.
[1] https://www.microsoft.com/en-us/research/uploads/prod/2023/0...
You would still read the exact same data you write, even if part of the process is not deterministic.
In 200 years the goal will be the same, and hopefully we will have a better algorithm of eliminating the noise.
Like a CD?
Perhaps such a device would be named a "Concise Disc".
I am assuming that you can store any arbitrary data on Azure, including music
Obviously the less overt signalling you can get away with, the better the joke is for those who get it. But if you provide so little that people confuse the joke for genuine ignorance, the joke isn't working as well as it could. I'm not the official joke arbiter by any means, but I did read your comment and think about whether you intended this specific joke, and concluded that you probably didn't.
If you genuinely weren't aware of CDs, that's fine, see https://xkcd.com/1053/.
Microsoft made ransomware attacks easier. Now they are redefining "easier" to harder. /s
indestructible, glass, pick one.