Platt Perspective on Business and Technology

Usage challenges that drive the evolution and growth of information technology – 3

Posted in business and convergent technologies by Timothy Platt on January 14, 2012

This is my third installment in a series on information technology innovation and on how it is driven by pressures on both the technology and user sides (see Part 1 and Part 2.)

Part 1 of this series was in large part about bioinformatics and the explosive growth of its information volume as that would drive change in basic internet technology and infrastructure, and Part 2 picked up on that working test case example. Basically, we have just seen the beginning of this as a stress test demand on the internet and on its backbone architecture and scalability. And bioinformatics simply represents one example of a large and growing number of applications where information volumes that would be processed and shared online, are and will continue to grow in a seemingly open-ended manner. And as I noted at the beginning of Part 2:

• When you sufficiently change any system quantitatively you reach a point where you of necessity change it qualitatively too.
• Put somewhat differently but to the same point, linear scalability is always limited for any real world system – including those designed to be linearly scalable.

My objective in this posting is to switch from discussion of networked flow and sharing of this increasing volume of data to look into some of the issues and challenges that arise in processing and managing it in specific computer systems. And I note here, that as computational speed and complexity increase, qualitative-change pressures enter in there too. And one effect of that has to be a blurring of the distinctions between local and networked in carrying this computational load. I said at the end of Part 2 that I would discuss the challenges of achieving exascale computing and argue that this is a turning point where the boundaries between local, single site computation and distributed and networked computation may in effect disappear and for many systems. And the basic definition of network per se and of what is assumed in that term will have to change accordingly.

My starting point for that can be found in an initiative called Internet2 – a strictly US based (as of this writing) consortium of member organizations drawn from research and education, industry and government that seeks to develop the standards and technology for a new, second generation internet backbone, and with equally advanced innovations into connecting into and utilizing this new system.

Efficiently scaling internet traffic volume up beyond the point where core internet technology up to now has been efficient, and beyond where it can be linearly scalable with that older technology alone is just one of its goals. But that is a core objective of Internet2. In this, a goal of Internet2 is to create new and hopefully greatly expanded ranges in scale and information flow-volume where linear scalability, without further fundamental redesign, is achievable. Think of emerging data-intensive applications and needs such as provided by an increasingly data-flow intensive bioinformatics information marketplace as constituting a driver of change, demanding that effective Internet2 and related approaches be developed and implemented. And with that, I switch directions in this discussion to turn explicitly to exascale computing itself.

As of this writing, the United States government has earmarked $126 million to go into the development of exascale computing systems for 2012 alone, and that is just from areas of the national budget that are publically announced. Funds directed towards this initiative from classified sources such as the National Security Agency (NSA) would not be included in that number, and in all likelihood a true measure of US government investment in this initiative would be at least twice the more publically stated level. And with that unlisted but likely participant in mind I note in passing that another data volume-intensive problem and application area of interest has to be in data encryption, and in both securely encrypting sensitive information and in code breaking to reveal information content of others. An increase in computational speeds achievable from the existing petascale computing standard to a true exascale standard would render a lot of archived but up to now unreadable information traffic quite readable, and it would move a lot of what can be decoded now, but with delays into being real-time available. But I just note that in passing, to suggest that $126 million is a low-ball estimate of any true number, and probably by a large margin.

A great many private sector participants are investing in this too, and from both the hardware and software sides. In this, weather forecasting, oil and natural gas exploration, rational drug design in pharmaceutical research and a host of other exascale problems are demanding this investment, if these organizations are to remain relevant.

• And this is where the boundaries between local and network blur, but with a very important, fundamental restriction imposed on the possible – the speed of light.

Single core central processing units (CPU) can only process so much information at any one time, even if the absolute limit for what they can do has not been reached. This is because packing more and more computational capability into a single core requires making the components within it physically smaller and smaller – and the fact that individual atoms have a positive non-zero volume means a point is reached where component scale is reduced to a point where small numbers of atoms are involved in any given chip component. And quantum mechanical effects start to dominate the physics of these systems as this micro-scale and even nano-scale.

This includes, among other crucially important emergent properties, the way in which even a slight narrowing in a pathway in such a circuit becomes a break in that circuit, as atoms literally flow out of it, and to either side of the narrowing. With increasing granularity in the materials used as atomic scale is more closely approached and with geometric scale increases in the number of places in a circuit that narrowing can occur, this becomes an increasingly irksome problem with increases in production cost from increased numbers of unit failures, or at least in their reduction of efficiency for units produced, as well as in the production of units per se.

I am going to stop this posting at this point and continue it tomorrow, where I will look into multi-core processors, parallel computing and other issues that go into increasing computation throughput. Meanwhile, you can find this posting and series, and related postings at Ubiquitous Computing and Communications – everywhere all the time.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.

%d bloggers like this: