[MUSIC] We're facing a real data dilemma. We are creating a lot
more than we can store. What we do save is degraded
with each passing generation. And the networking technology that we
rely on uses way too much power. What if we could store every bit of our data on a media that is virtually indestructible that can scale in
ways that use a lot less energy? [MUSIC] >> That's the vision
that's driving us at Microsoft to radically rethink and invent new disruptive
technologies in three key areas: storage, networking and compute. [MUSIC] >> Over a third of
all the silent movies ever made no longer exist. Perhaps, you've heard of the
Universal Studios fire of 2008, when tens of thousands of master recordings of our
greatest artists were lost. So the media industry tell
us that the films, music, the content that they produce is
the value of their companies. And they go to extreme lengths
to try to maintain it. And yet, they acknowledge, even
with their best efforts, everything is slowly
degrading over time. And we hear that from
other industries too. Like the banking sector. They understand exactly how
risky current storage can be, how everything has to be backed up on a systematic basis and the
potential loss of critical data, that's enough to keep
them up at night. On a more personal note, think of all the precious data that
you carry around on your phone. All of that is just a mishap
away from being lost. This is going to change
all that forever. This is a platter of silica. Basically, it's a bit of glass and it probably
looks invisible to you. But if you look at it very carefully, you can see data that
is stored inside it. If you think of the hard
disk drive industry, it took them over 40 years to
reach their platter capacity. It took the optical
disk industry over 25 years to reach the capacity
that they're at today. We've been pushing on
this technology for only three years and already,
in this piece of glass we can store more data than
you store in an optical disk. And we're only just at the beginning. A piece of glass like
this theoretically can store hundreds of terabytes of data. At Microsoft, we've been
working hard at optimizing our cloud-scale storage using existing technologies like hard
disk drives and like tape. But we got to the stage where we just can't push it
as far as we want, and that led us to go out
and look for a new media. Our search led us to
Southampton University. They were writing data into glass
using pulsed femtosecond lasers. These are lasers that are firing
extremely short pulses of light into the piece of glass, and where that
laser heated the glass, a small dot was created and
that dot can store data. We call that a voxel. Think of it like an iceberg, it's got depth and it's
got orientation and we're able to control
both the orientation and the depth of the iceberg. And that's how we can store multiple
bits in a single voxel. We just thought this was
amazing, and we thought, how can we tame this technology
to provide a storage system which doesn't require
us to keep copying data from one generation to the next? It turns out that glass
is remarkably resilient. You can boil this in water. You can scrub it with steel wool. You can subject it to
electric magnetic pulses. You can even broil this at
500 degrees Fahrenheit. The data, once it's
written inside it, is going to survive all those things. It is truly one of the most resilient media
that has ever been found. Once you store your data in this, the date is going to
be there for 100, 1,000, even 10,000 years. And if you no longer want to keep
your data, it's really simple: you can just melt it down
and it becomes more glass. [MUSIC] >> We've just heard
about data storage as one of the challenges when
we're thinking about the future. The other one is
sustainable networking. It's about the energy and
the speed with which we can transfer data over the network
to access that storage. And in our team, we're working
very hard and looking at new optical technologies that we can bring to bear on optical networking. If you think of a traditional
data center at the moment, when you send some data, it goes to a server in a rack
to a top-of-rack switch. That top-of-rack switch
converts the data from electrons to photons and
sends it to another switch where it's converted back, so this can then decide
where to send it next. And this process is then repeated millions of times
across the datacenter. That means we're wasting huge amounts of energy doing these conversions. So what we're hoping
to do is allow you to convert it just once and then we're going to send it across
the entire network as light. One way to think about all this
is to think of this prism. If you shine sunlight into it, it's going to split the sunlight
up into different colors. And this property is going to allow
us to steer where the data goes. But the challenge is to
do that really fast. When I say really fast, I mean light-speed fast. So we went all in to
create an optical chip that could switch between the different light colors
at the nanosecond range. As for sustainability, every
time we convert from photons to electrons and electrons to
photons, it takes energy. So the great thing about
this new optical networking is that at the core it's
completely passive. All that happens is is that it
steers beams have lights. It requires no additional
energy to do that. We're trying to also
incorporate it with optical storage to
create new types of data centers that we can deploy in the most arid or ice-bound regions, out at sea or even underwater. [MUSIC] >>Most recently at Microsoft, we've been applying the same kind
Of radical thinking to compute. There are a lot of optical
technologies like 4K and 8K cameras, like SLMs which are
basically screens. They all becoming mainstream now, and we're beginning to ask, are the things that are being
explored in the past that might make sense now when we have
these newer components? One of the things that
we're really interested in doing is looking at
whether we can create new machine learning
algorithms that can be designed from the ground-up to work with the kind of operations that you can do easily in the
optical compute world. We're seeing a future
where we can bring all of these technologies together. We're not there yet, but every day we're making
progress on that dream. [MUSIC] >> If you think of all the
things that we've lost to time, the Library of Alexandria, the Manuscripts of da Vinci, irreplaceable writings and
artistic masterpieces. Imagine if we could
promise for you to store anything that's
important to you. Your e-mails, your documents,
your photos – forever. For me, on here I'm
storing my wedding photos. So my great, great, great, great, great, great grandchildren
can enjoy them. It's really quite
powerful to be able to store your information forever. It's going to change the way
that we look at the past. It's going to change the way that
we collaborate in the future. It is literally going to transform
existence as we know it. Now, how powerful is that? [MUSIC]
Digital long-term cold storage backups have been a concern for some time. Magnetic tape storage is still used a lot but tape technology has limits, and warehouses full of tapes are unsustainable long term. This new technology provides a way to increase the amount of data storage per mm, lowering the cost of the physical space the medium used takes up in the storage vaults in which they are kept.
Seems this is just for archiving purposes (WORM drive), single-use writes like the old CD-Rs.
You can use it for scheduled backups where it's a capture in time, or a finished project in its entirety you may want to revisit later.
A write-once memory can usually invalidate previously written data. So, even though the old data may be there, it can be invalidated and then a new version written. One way is to just keep a table of invalid previously-written data blocks.
But, this is not intended for real-time online read/write. It is only for long-term storage.
Good question, but if the storage is dense enough and cheap enough, WGAF? They're not taking about RAM and short term storage, so whilst archive is a natural starting point for this, if you have a series of wafers each containing a load of medium dynamic data, your emails or whatever (you'll get new ones, some will be deleted, most will just hand around for a decade or two), you can store them and ignore the deleted ones. Once the wafer has more deleted than live sectors, you defrag by copying the good stuff to the next wafer and melting the old one down for reuse. Write once isn't an issue if the medium is so cheap you don't care - look at home burn cd's for transporting a couple of Word documents, pretty common in it's day.