Exploring The Machine: A Q&A with Martin Fink

November 4, 2014 • Blog Post • BY HPE MATTER STAFF

Thank you to everyone who joined our live event with Martin Fink on The Machine. For those of you who weren't able to attend, you can watch the full replay here. We received quite a few questions from the audience that Martin didn't have time to address during the event, so we sat down with him afterwards to get his thoughts.

What are the implications of using Memristors for both storage and computation?

We've talked in the past about how Memristors might be used to produce logic gates and even synapse-like behavior, but for The Machine were using Memristors "just" for storage. This is still pretty revolutionary: were going to collapse main memory (typically DRAM) and storage (Flash and HDDs) into a single tier comprised of massive amounts of Memristor memory.

You can get more detail about Universal Memory here, but the short version is by the end of the decade, we believe the shift to universal, non-volatile memory is inevitable if were to keep up with the data explosion without running out of power. In fact, the data/power matrix will make conventional technology impossible for all practical use cases. Replacing the existing memory hierarchy with universal memory enables a massive simplification in programming. Virtually every piece of software ever written is suddenly hopelessly inefficient – millions of lines of code do nothing but cope with and mask the overhead of shuffling data between tiers of memory and storage.

How might The Machine be applied in the consumer market?

The data explosion we're facing will affect every aspect of the technology market, both for consumer and enterprise customers. The first instantiation of The Machine will probably be as a service – i.e. you interact with it via the web, cloud-style, rather than buying a box. It will subsequently be delivered as a system – at the scale of a handheld, server, or data center. Then, it'll get built into all sorts of smart devices – electricity meters, cell towers, cell phones, jet engines – the list is endless. Some of those will be consumer devices, some not.

Were designing The Machine architecture to be almost infinitely scalable, so the possibilities are endless.

Why has HP chosen to announce The Machine as early as now when it will not be released anytime soon? Doesn't this reveal valuable information to competitors?

There are a couple of parts to this. Firstly, as we've discussed, we're headed toward an inflection point where the massive influx of data can't be handled by whats currently on the market. Weve discussed our concept for solving the problem with select customers, and we've had a universally positive response. Our most forward-looking customers get it, and want it, and HP can provide them a roadmap to the future.

Secondly, multiple components of The Machine are well on their way. You'll continue to see news about The Machine over the next few years as a concrete timeline comes into place. In addition, we're intending for the new OS to be Open Source and were going to be making programming tools and emulators available. The more time we can give the developer community to think about and start to take advantage of the new capabilities that The Machine will enable, the better.

What would happen to The Machine if HP is splitting?

We aren't expecting the split to have a significant effect on The Machine project. It's not as if the two new companies will be forbidden to talk to each other!

How do you distinguish what you are doing with The Machine from what IBM and Watson?

I'm glad you asked, because this is important. The Machine is not like IBM's Watson.

Watson tries to emulate human cognition by combining three existing technologies into one: natural language processing, hypothesis generation and evaluation, and machine learning.

The Machine is a completely different entity from Watson: it does not attempt to emulate human cognition. Rather, its purpose is to revolutionize the way data is processed in very large business and analytical applications. The Machine is a parallel computing device capable of storing enormous amounts of data in-non-volatile memory close to the processing units, and with extremely low latencies in data transfers within the machine. The Machine thus acts a very dense cluster where network latency and machine to machine data shuffling disappear.

Also, as a practical matter, Watson is a powerful, but conventional computer that needs lots of space and power. The Machine is a new type of inherently power-efficient computer that scales from handheld to supercomputer.

What is HP doing to make sure that applications are ready to reside on the new OS and take advantage of the new architecture when The Machine is introduced?

The Machine is primarily aimed at new applications, particularly involving datasets that are too large to hold in memory today (even if you have millions of dollars and your own personal power station). If you'd like more on this, you might like this article on Graph Analytics I did for HP Matter recently.

We want you – customers, partners and the entire developer community – to get your hands on the software building blocks as soon as possible. We want you to start kicking the tires and starting to explore possibilities we haven't even thought of. You'll be hearing more from us on this topic very soon.

Having said that, there are existing business applications that can extract significant benefit from The Machine. That's why the operating system for The Machine has two phases: first we will use familiar programming constructs to allow legacy applications to run with improved performance. The second phase will build on this to deliver the first all-new OS in decades, enabling developers to achieve orders-of-magnitude improvements in speed and scale from a whole new class of applications.

What problems or challenges are there to the development of The Machine that might prevent its success?

As with all big research projects, there will always be challenges to be overcome. What I will say is this: The Machine isnt an all-or-nothing project. In addition to The Machine as a finished range of projects, each of the component technologies is interesting on its own: Memristor as drop-in DRAM and Flash replacements. Photonic interconnects to make data centers more manageable and energy efficient. Management-at-scale techniques to multiply the effectiveness of sysadmins. The list goes on.

Think of it this way: even if Apollo hadn't landed on the moon, the space program would still have given us microchips and Velcro!

What is the production status of Memristors?

We've demonstrated that Memristor works, and weve made many test chips to explore different materials and layouts. Were in the commercialization phase now with our joint development partner SK Hynix, and we're on track to begin sampling chips to our partners in 2016.

If I run a F500 caliber IT infrastructure today, what challenges will I face in "migrating" to The Machine?

Our job at HP Labs is to look far ahead into the future, to identify and address the problems we know our customers are going to be facing. So in a sense, the question is backwards: The Machine is designed to overcome the challenges we know are coming.

We're ready to help our customers meet their current challenges and run their businesses today, while helping them evolve along with us as we transition into world of The Machine.

For regulatory and compliance reasons, how do I archive and back up data in this environment?

HP offers a range of products and services today to automate all these processes and give our customers peace of mind that their data is safe and compliant with all local regulatory requirements. That outcome won't change, although the underlying technology may be very different.

Can you really get software developers interested before you prove the hardware works? I think not.

Were very much hoping to prove you wrong! We've talked to hundreds of customers and partners about this. Honestly, we were expecting many to agree with you. To our surprise, the reaction can be summarized as its about time!

As you may know, we're intending for the new OS to be Open Source. We're going to be making programming tools and emulators available so that the entire developer world can contribute and start to take advantage of the new capabilities that The Machine will enable. I hope you'll join us.




Click to learn how The Machine will advance four emergent technologies in parallel to prevent the rising data flow from flooding conventional IT systems and to disrupt the way we think about computing.