• The Internet and the World Wide Web as we know them today
• Search technology - Where once we filed, today we search
• Cluster computing
• The transformation of science via computation

In this post, we summarize just a sample of your additions (we have grabbed text from your posted comments, without a lot of editing, so this will be loose - “it’s the thoughts that count”) and invite your further comments - cleaning up these additions, or providing others.  Please let us hear from you!

Secure communication - the foundation of e-commerce

All of e-commerce relies on the results of computing research:  the Internet, the World Wide Web, cluster computing, parallel relational database systems, cryptography and algorithms for secure credit card transactions.  Here, we focus on the latter.  Without secure communication - for example, the ability to conduct a credit card transaction with an online merchant - there would be no e-commerce.  The complex of events (both theoretical advances and deployment of practical, useful software) that allow a user to type a credit card number into a web browser and be reasonably assured of its safety is a game-changer, making secure communication and secure commerce a reality for (potentially) all users of the Internet.  Without these artifacts, we would have no Amazon.com, no eBay, no thriving online pornography industry, …

Mobile computing and communication

Twenty years ago, computing was a desktop experience.  “Portable computers” were the size of a briefcase.  Communication was via 9600 baud telephone modem.  Contrast that to today:  2 pound laptops that fit in a mailing envelope, mobile phones with Web browsers that fit in a shirt pocket, and ubiquitous WiFi and 3G cellular at many millions of bits per second.   Clearly, mobile computing and communication - the untethered lifestyle - is a game-changer.

Expert systems become ubiquitous

Thousands of routine decisions daily are made by computer systems that have specialized knowledge of a problem area. In the past, rule changes at a central office - e.g., the IRS, or the headquarters for a corporation - were incorporated slowly into practice. With expert systems, the people making the decisions have the benefit of codified knowledge bases that reflect current policy and practices.

Research on expert systems began in the 1970’s with support from DARPA, the National Institutes of Heath, and NSF. Expert systems have subsequently become an essential part of the IT toolkit for every major company. Help desks, credit checking and equipment troubleshooting are examples of systems that have been replicated many times over and are routinely saving money for business and public institutions.

Expert systems technology is a game-changer.

Robotics in everyday life

Twenty years ago, robots appeared in artificial intelligence laboratories, automated assembly lines, and science fiction movies.  In recent years, iRobot Corporation has sold roughly 1,000,000 Roomba robotic home vacuum cleaners annually, and multiple robotic automobiles have completed the DARPA Grand Challenge and Urban Challenge, autonomously navigating a 150-mile desert course and a 60-mile urban course.  Robots have entered the mainstream of society, integrating a wide variety of Artificial Intelligence technologies such as computer vision, sensing, and planning.  This is a game-changer, and the best clearly is yet to come.

Digital media

Today, almost no one thinks of photography in any form other than digital.  The means by which we capture, edit, and share digital images are the result of multiple breakthroughs in computer science.

Similarly, digital compact disc audio - a breakthrough when it entered the mainstream only two dozen years ago - is going the way of the dinosaur, replaced by MP3 audio on personal devices such as iPods.

Our video entertainment is in digital form too - whether on a DVD, a personal video device, streaming media, or a video game.

Digital media is revolutionizing entertainment and the entertainment industry - a game-changer.

GPS, mapping, and navigation

GPS - the ability to pinpoint your position nearly anywhere on earth - is a marvel.  But even more amazing are the algorithms that provide navigation - available on the Web, and in $200 self-contained portable devices from Garmin, TomTom, and others.  GPS, mapping, and navigation are game-changers.

Collaborative filtering and recommender systems

Collaborative filtering and recommender systems dramatically altered how we think about computing applications by introducing the idea that the actions and preferences of other people could be a useful resource in computations intended to support someone else’s activities.  This is easily appreciated by a broad audience - anyone who has used Amazon.com’s “people who bought this also bought…” or other social features; a somewhat narrower audience will also appreciate that a major improvement in search engine performance occurred when they started taking into account link structures and then click behaviors.

There’s a clear  tie to computing research, both in work on algorithms for using data from other people, and in interfaces for collecting it and presenting predictions or recommendations.  The idea was first articulated in CACM and in the ACM CSCW and CHI conferences, and there are now thousands of papers about it.

A few additional ideas that were suggested

These need fleshing out or weeding out!  Our comments in [blue brackets] …

• Something related to applications of machine learning - the applications within computing (e.g., NLP, vision, graphics), to other sciences (with big data), to finance (credit card fraud, and dare I say Wall Street) abound [for sure - needs fleshing out]
• Something related to advances in software engineering, and the application of logic to analyzing both hardware and software designs and artifacts [the application of logic might work; we still have a "software crisis," though, and "there (still) is no silver bullet," so need to be careful with claims]
• Something related to scientific computing and large-scale computational science, simulations, etc. [we meant this to be covered by one of our original topics - "the transformation of science via computation"]
• Virtualization [can someone say "1960s"?]
• Network coding [would need to be painted larger]
• Compressed sampling/sensing [would need to be painted larger]
• Quantum computing [premature]
• Elliptic curve crypto [covered crypto under secure communication]
• Molecular computing [come see us in 10 years!]
• Randomized algorithms [would need to be painted larger - colored with applications]
• Theory of distributed computing: impossibility results, Byzantine generals [we meant to feature this under our "cluster computing" topic, which relies integrally on these algorithms; cluster computing is not a hardware breakthrough, it's a distributed algorithms breakthrough!]
• Wearable/ubiquitous/mobile computing [covered under mobile computing and communication, a new topic above]
• Sensor networks [tell me more]
• Human computation (Captchas, the ESP game, etc.) [maybe ...]
• Computational microeconomics: ad placement, automated mechanism design [sounds good - say more!]

Again, we invite your comments!  Let us hear from you!

– Ed Lazowska and Peter Lee

–

The CCC blog has published a couple of articles on the multi-core challenge, all emphasizing the difficulty of making parallel programming prevalent and, hence, the difficulty of leveraging multi-core systems in mass markets. The challenge is, indeed, significant and requires important investments in research and development; but, at UPCRC Illinois, we do not believe that the sky is falling.

Parallel programming, as currently practiced, is hard: Programs, especially shared memory programs, are prone to subtle, hard-to-find synchronization bugs and parallel performance is elusive. One can reach two possible conclusions from this situation: It is possible that parallel programming is inherently hard, in which case, indeed the sky is falling. An alternative view is that, intrinsically, parallel programming is not significantly harder than sequential programming; rather, it is hampered by the lack of adequate languages, tools and architectures.  In this alternative view, different practices, supported by the right infrastructure, can make parallel programming prevalent.

This alternative, optimistic view is based on many years of experience with parallel programming. While some concurrent code, e.g., OS code, is often hard to write and debug, there are many forms of parallelism that are relatively easy to master: Many parallel scientific codes are written by scientists with limited CS education; the time spent handling parallelism is a small fraction of the time spent developing a large parallel scientific code. Parallelism can be hidden behind an SQL interface and exploited by programmers with little difficulty. Many programmers develop GUI’s that are, in effect, parallel programs, using specialized frameworks. Parallelism can be exposed using scripted object systems such as Squeak Etoys in ways that enable young children to write parallel programs. These examples seem to indicate that it is not parallelism per se that is hard to handle; rather it is the unstructured, unconstrained interaction between concurrent threads that result in code that is hard to understand both from a correctness and performance view, hence hard to debug and tune.

The state-of-the-art in parallel programming is what sequential computing was several decades ago. A major reason for this situation is that parallel programming has been an art exercised by a group of experts whose small population did not justify major investments in programming environments aimed at making their life easier. This reason disappears as parallelism becomes available on all platforms. Furthermore, we can make faster progress now because we understand well the principles it takes to make programming easier — principles such as safety, encapsulation, modularity, or separation of concerns; we also have more experience in developing sophisticated IDE’s.

What will it take to bring these principles of computer science to parallel programming? It will require a broad based attack across the system stack. As has been said in these blogs, we need research in languages, compilers, runtime, libraries, tools, hardware … What has not been said explicitly is that none of these areas are likely to produce the silver bullet on their own. The solution that will work eventually will be one that brings together technologies from all these areas to bear on each other. However, we do not have the luxury of doing this via incremental and reactive changes over decades. The research truly needs to be interdisciplinary and the idea of co-design needs to be internalized. Unfortunately, the mainstream systems community has all but abandoned this mode of research in the last several years. Language researchers are locked into mechanisms that will only be supported by commodity hardware and hardware researchers are locked into a mode that requires supporting the lowest common denominator software. It is imperative that we break out of these shells and get the research community into a mindset that we are truly looking to define a new age of computing — a mindset that nurtures research where a clean system slate is an acceptable starting point.

The sky is not falling, but the ground is shifting rapidly. The multi-core challenge requires a concerted effort of academia and industry to generate new capabilities. We are confident that in the future, as in the past, new capabilities will breed new applications. Multi-core parallelism can be leveraged to develop human-centered consumer products that provide more intelligent and more intuitive interfaces through better graphics and vision, better speech and text processing and better modeling of the user and the environment.

The task of providing better performance is shifting from the hardware to the software. This is an exciting time for Computer Science.

Marc Snir
4323 Siebel Center, 201 N Goodwin, IL 61801
Tel (217) 244 6568
Web http://www.cs.uiuc.edu/homes/snir

The workshop vastly exceeded my expectations – 8 hours of brainstorming about strategies and best practices, in four areas:

·     “Go Outside Your Box” – what strategies can we adopt to increase collaboration across subfields and with other fields?

·     “The World Needs Us” – how to contribute to the solution of societal “Grand Challenge” problems while simultaneously driving computing research forward.

·     “Breaking the Cycle” – can we change the reward structure to decrease incrementalism, encouraging long-range thinking?

·     “Serving the Community” – how can we further increase the culture of service to the research community and to the nation?

I’m sure others will blog on various aspects, and teams have formed to write up specific strategies and best practices.  But here are a few things that really struck me:

Computer science:  the ever-expanding sphere

A model of how our field evolves can help us make smart decisions.  Think of computer science as an ever-expanding sphere (this analogy is due to Alfred Spector; all graphics are due to Peter Lee).  We transform other fields and we change the world.  We do this not just through the application of computation, but through the introduction of computational thinking.  When we transform these fields, we make new discoveries about our own field that enlarge our “bag of tricks” – our ability to transform other fields.  So we constantly reinvent ourselves by reinventing others.

We’ve transformed circuit design, publishing, photography, communication, mechanical CAD, certain fields of science, …  We’re in the process of transforming biology, transportation, …  And we’re always transforming ourselves.  Computer science truly is an endless frontier.

What this means is that even when working inside the sphere, we’ve got to be looking outward.  And at the edges of the sphere, we’ve got to be embracing others, because that’s how we reinvent ourselves.

Computer science lives in Pasteur’s Quadrant

The vast majority of work in our field is motivated both by concerns of use and by a desire to evolve principles of enduring value.  If anything, we may be too much in Pasteur’s Quadrant – we may place too little value on research without obvious utility, and we may be too reluctant to reject as “not computer science” work that’s focused on applications where it may not be obvious that our own field will be advanced.  (Jim Gray had the stature and courage to pioneer our move into data-intensive eScience – today, the transformations this has stimulated “within the sphere” are obvious.)

Lots of the action is at the interfaces

This is true fractally – it’s true of the interfaces between computer science and other fields (the edges of the sphere), and it’s true of the interfaces between subfields of computer science.

We’ve got to produce students who are comfortable at these interfaces.  It’s increasingly difficult.  Stuart Russell observed that “Bohr drives Pasteur” – we need strength at the core, and the core is ever-expanding.  At the same time, students need to be able to make connections.  I’m concerned we’re making the wrong tradeoffs these days.  Students enter graduate school with records that look like promotion cases a decade ago!  We decrease course requirements to get students engaged in our own research as quickly as possible.  Our colloquia are half-empty because everyone’s too busy beavering away to attend.  These factors decrease breadth and agility within the sphere.  We don’t require minors, which would expose students to other fields – this decreases the ability to work at the edge of the sphere.  As a field, we should tackle these issues head-on.

Visions, incremental progress, and random walks

A research project needs to be hard enough to be interesting, and easy enough to be doable.   There needs to be a vision – a sense of where you and your colleagues are headed over a five-year or ten-year period.  And it needs to be tackled in what Alfred Spector calls “factorizable pieces.”

If there’s a vision, then a “factorizable piece” may appear incremental, but it’s headed somewhere important.  Without a vision, it’s part of a random walk.  It’s important to differentiate these!  A goal of the CCC “visioning workshop” process is to articulate some of these visions for our field.

Bob Sproull pointed us to a wonderful paper by Ivan Sutherland on the conduct of research – “Technology and Courage.”  Read it!

Onward

I’m sure others will blog on various aspects of the workshop.  Look in the mirror – is there a field you’d rather be part of?

– Ed Lazowska

• The advance needs to be “game changing,” in the sense of dramatically altering how we think about computing and its applications.
• The importance of the advance needs to be obvious and easily appreciated by a wide audience.
• There needs to be a clear tie to computing research (or to infrastructure initiatives that build upon research and were sponsored by computing research organizations).
• We’re particularly interested in highlighting the impact of federally-funded university-based research.

We’re focusing on work carried out in the past 20 years or so, in part because of the upcoming 20-year celebrations for the CISE directorate at NSF. Of course, lots of great fundamental research can take more than 20 years before the impact becomes obvious, but even in such cases there is usually continuing influences on more recent research that can be cited here.

To get your juices flowing, here are four game-changers that we definitely think belong on the list. Use these to think about others that belong on the list, or feel free to argue with our choices.

The Internet and the World Wide Web as we know them today

In 1988 — 20 years ago — ARPANET became NSFNET. At the time, there were only about 50,000 hosts spread across only about 150 networks. In 1989, CNRI connected MCImail to the Internet — the first “commercial use.” In 1992, NCSA Mosaic triggered the explosive growth of the World Wide Web. In 1995, full commercialization of the Internet was achieved, with roughly 6,000,000 hosts spread across roughly 50,000 networks. Today, there are more than half a billion Internet hosts, and an estimated 1.5 billion Internet users.

While many of the underlying technologies (digital packet switching, ARPANET, TCP/IP) predate the 20-year window, the transition from the relatively closed ARPANET to the totally open Internet and World Wide Web as we know them today falls squarely within that window. NSF-supported contributions included CSnet, NSFNET, and NCSA Mosaic.

The Internet and the World Wide Web are game-changers.

Where once we filed, today we search

The vast majority of the world’s information is available online today, and we find what we need — whether across the continent or on our own personal computer — by searching, rather than by organizing the information for later retrieval.

Research on the retrieval of unstructured information is based on decades of fundamental research in both computer science theory and AI. But the paradigm shift that is web crawling and indexing and desktop search is much more recent. It traces its roots to university projects such as WebCrawler, MetaCrawler, Lycos, Excite, Inktomi, and the NSF Digital Libraries Initiative research which begat Google.

Search is a game-changer.

Cluster computing

At the risk of offending our many computer architect friends, we’re going to assert that cluster computing is the most significant advance in computer architecture in the past 20 years.

A decade ago, Jeff Bezos was featured in magazine advertisements for the DEC AlphaServer, because that’s what Amazon.com ran on — the biggest shared-memory multiprocessor that could be built. Similarly, the AltaVista search engine was designed to showcase the capabilities of big SMP’s with 64-bit addressing.

Today, this seems laughable. Companies such as Google and Amazon.com replicate and partition applications across clusters of tens of thousands of cheap commodity single-board computers, using a variety of software techniques to achieve reliability, availability, and scalability.

The notion of hardware “bricks” probably can be traced to Inktomi, a byproduct of the Berkeley Networks of Workstations project. The software techniques are drawn from several decades of research on distributed algorithms.

Cluster computing is a game-changer.

The transformation of science via computation

The traditional three legs of the scientific stool are theory, experimentation, and observation. In the past 20 years, computer simulation has joined these as a fundamental approach to science, driven largely by the NSF Supercomputer Centers and PACI programs. Entire branches of physics, chemistry, astronomy, and other fields have been transformed.

Today, a second transformation is underway — a transformation to data-centered eScience, which requires semi-automated discovery in enormous volumes of data using techniques such as data mining and machine learning, much of which is based on years of basic research in statistics, optimization theory, and algorithms.

Computational science is a game-changer.

Some non-inclusions

Quantum computing. There is huge potential here, but the impact hasn’t been felt yet.

Simultaneous multithreading. We claim that this, and many other important advances in computer architecture, are dominated by cluster computing. (Remember, we’re trying to be provocative here! Blame Dave Ditzel, who put this idea into Ed’s head.)

Your part goes here!

What’s your reaction to the four game-changers that we’ve identified? Do you agree that they belong on the list? If not, why not? If so, what do you think were the principal components of each — the key contributing research results?

Even more importantly, give us eight more! What are your nominees for game-changing advances from computing research conducted in the past 20 years?

Give us your thoughts!

– Ed Lazowska and Peter Lee

These highlights are designed to provide easily digestible, compelling nuggets of computing research work. Members of Congress, the Administration, and funding agency managers and directors are some of the main audiences for these web pages. We believe the highlights should also prove to be useful for the entire research community. The highlights can be accessed directly, received by email, RSS feed, or even embedded in your own web page.

The current Computing Research Highlight of the Week describes a new algorithm from UCSD researchers that performs route computation in a way that may lead to major improvements in network efficiency. Check it out — it is punchy, informative, and makes good use of some simple graphics while at the same time providing links to the scientific publication and full press release.

So, please submit your own highlights! The response thus far has been very good, and we expect that many people outside of our community, including key decision makers, will make good use of the information.

• The Industrial Revolution of Data. Today’s world-wide web remains a staggering tribute to the typing abilities of the human race. But even with a growing global population, typists are not a scalable source of bit-production going forward. We are entering an era where the overwhelming majority of information will not be hand-crafted. It will be stamped out by machines: software logs, cameras, microphones, GPS transceivers, sensor networks, RFID readers, and so on. This is inevitable. It has already begun to change the computing marketplace: most organizations of size now realize they can afford to save and mine all their logs, and are looking for inexpensive ways to do so. The startup world has responded with a flurry of parallel database and data analytics companies.
• The Crisis of the Three C’s: Coders, Clouds and Cores. Meanwhile, it’s no news that software development is far, far too difficult. In his Turing Award talk a decade ago, Jim Gray identified radical improvements in programming among his 13 remaining long-term challenges for computing — alongside passing the Turing test and building Vannevar Bush’s Memex. What’s changed on this front since 1998 is the rapid rise of parallelism that my colleagues have been blogging about here. Cloud computing infrastructure, with its “shared-nothing” clusters of machines, demands parallel and distributed programs today. Manycore architectures will demand parallelism at a finer grain in the next few years. The pressing need for parallel software — and armies of fluent software developers to build it — raises both the difficulty and the stakes of the Grand Challenge that Jim Gray highlighted in 1998.

Given this background, what excites me these days is that the trend may bring some new solutions to the crisis, in a surprisingly organic way. 

For over twenty years, “Big Data” has been a sustained bright spot in parallel computing. SQL has been a successful, massively parallel programming language since the late 1980’s, when Teradata (a survivor that evaded Dave Patterson’s Rolls of the Dead) first commercialized parallel database research from projects like Gamma and Bubba. In recent years, SQL has been joined by Google’s MapReduce framework, which is bringing algorithmicists into massive data processing in a way that SQL never did. Both SQL and MapReduce will likely thrive, and may well converge: two parallel database startup companies recently announcedintegrated implementations of SQL and MapReduce. (Full disclosure: I advise Greenplum, one of those companies.)

SQL and MapReduce programmers do not think much about parallelism. Rather than trying to unravel an algorithm into separate threads, they focus on chopping up sets of input data into pieces, which get pumped through copies of a single sequential program running asynchronously on each processor. In parallel programming jargon, this kind of code is sometimes dismissively referred to as being “embarrassingly parallel”. But very often, the simplest ideas are the most fertile. Programmers “get” these approaches to parallelism. And remember: the Coders are part of the Crisis of the Three C’s, and the key is to make lots of them happy and productive.

But can those programmers lead us anywhere interesting? The most intriguing part of this story is that in the last 5-10 years, the set-oriented, data-centric approach has been gaining footholds well outside of batch-oriented data parallelism. There has been a groundswell of work on “declarative”, data-centric languages for a variety of domain-specific tasks, mostly using extensions of Datalog. These languages have been popping up in networking and distributed systems, natural language processing, compiler analysis, modular robotics, security, and video games, among other applications. And they are being proposed for tasks that are not embarrassingly parallel. It turns out that focusing on the data can make a broad class of programs simpler — much simpler! — to express.

Networking and distributed coordination protocols are one good example. They run in parallel, and are themselves a key to cloud services. In our work on Declarative Networking over the last years, we showed that a wide range of network and distributed coordination protocols are remarkably easy to express in a data-centric language. For example, our version of the Chord Distributed Hash Table (DHT) protocol is 47 lines of our Overlog language; the reference implementation is over 10,000 lines of C++. (DHTs are a key component of cloud services like Amazon’s Dynamo.) Meanwhile, graduate students at Harvard prototyped a simple version of the tricky Paxos consensus protocol in an alpha edition of Overlog in 44 rules. (Paxos is a key component in cloud infrastructure like Google’s GFS file system and Chubby lock manager.) We have other examples where our declarative programs are line-for-line translations of pseudo-code from research papers. These are the kinds of scenarios where the quantitative differences are best captured qualitatively. You can print out our Chord implementation on one sheet of paper, take it down to the coffee shop, and figure it out. Doing that with 10,000 lines of C++ would be a superhuman feat of Programmer-Fu, and a big waste of paper.

Machine Learning is another area where data-centric declarative programming seems to help with parallelism and distribution. A group at Stanford pointed to a range of standard machine learning tasks that can be expressed almost trivially as MapReduce programs, without any requirement for parallel programming expertise. More deeply, a number of the fundamental algorithms driving Machine Learning center on “message-passing” algorithms like Belief Propagation and Junction Trees that work on a computational model of explicit dataflow, rather than shared memory. Research in ML over sensornets showed how to overlay that logical communication onto a physical network. And these inference networks — much like DHTs — turn out to be a good fit to Overlog. (Distributed Junction Trees in 39 rules, anyone?) The Dyna language is another good example, with a focus on (currently single-node) Natural Language Processing.

I’ll be the first to admit that Datalog syntax is horrible, and it is not a reasonable language for developers. There is much to be done before adoption of complex data-centric languages can occur. But what excites me here is that the main positive trend in parallel programming — the one driven by the Industrial Revolution of Data, the one with programmer feet on the street — that trend feeds into this promising new generation of much richer data-centric languages. If MapReduce is the boot camp for a next round of parallel languages, those languages are likely to be data-centric. And there’s growing reason to believe that the data-centric approach will suit a wide range of tasks.

– Joe Hellerstein is a Professor of Computer Science at the University of California, Berkeley. His research focuses on data management and networking.

Thanks for the opportunity to update the community on what has been happening recently with the Network Science and Engineering (NetSE) effort, from my perspective as chair of the NetSE Council.

Let me explain my take on NetSE with an anecdote from my Georgia Tech colleague Mike Best based on a recent trip he made to Africa. Mike and his group met with a group of chiefs of the Acholi people in Northern Uganda. This is an area that has suffered through profound conflict and lacks for essentially any communication technology. Mike and his team wanted to engage in participatory design to understand the existing communication needs, unmet needs and requirements, and latent requirements.

They were very cautious not to influence the conversation towards modern communication technologies so they did not mention specific systems. But after about thirty minutes of this exercise one of the chiefs finally stated, “We want the internet. Unless you have something better.”

To me, NetSE is about the potential for something better. That isn’t to take away from how incredible the Internet is, but that success has led to a dependence on an infrastructure that we understand surprisingly little about. Figuring out what “better” means and how we might get there is a challenge that is intellectual, economic, political and social. In other words, hard, but incredibly important.

The last couple of months have been busy for the NetSE community. Five workshops and meetings have taken place since mid-June covering Network Design and X, where X has been Network Science, Societal Values, Theoretical Computer Science, Behavioral Economics, and Network Engineering. The goal of these activities has been to add to all the good work on research opportunities done under the auspices of GENI, but without the yoke of justifying a large facility.

NetSE is shaping up to be strongly disciplinary AND interdisciplinary. There remain major challenges and opportunities in the core disciplines of networking and distributed systems, as well as across disciplines in and out of CISE. For example, technology advances are producing the ability to program all the way down to the photon or RF wavelength. How can and should future networks take advantage of programmability at this extreme? In the interdisciplinary vein, there are important and exciting opportunities at the intersection of human behavior and network behavior. How should home networks be structured so that mere mortals can deploy and manage them?

Over the next couple of months, we will be synthesizing the output of the various activities into a NetSE research agenda that will include recommendations to funding agencies about what is needed to advance the agenda. You can watch for updates on the NetSE page hosted by the CCC at www.cra.org/ccc/netse.php.

– Ellen Zegura is Professor and Chair of Computer Science, School of Computer Science, College of Computing, at the Georgia Institute of Technology.

–

For over thirty years, we have watched the great cycle of innovation defined by the commodity hardware/software ecosystem — faster processors enable software with new features and capabilities that in turn require faster processors, which beget new software. The great wheel has turned, but it no more, as power constraints and device physics now limit the performance achievable with single microprocessors. Multicore chips — those with multiple, lower power processors per chip — are now the norm. Moreover, current multicore chips (those with 4-8 cores/chip) are but the beginning. We can expect hundreds of cores per chip in the future, with diverse functionality (graphics, packet protocol processing, DSP, cryptography and other features).

The software research challenge is clear — developing effective programming abstractions and tools that hide the diversity of multicore chips and features while exploiting their performance for important applications. Hence, we need a vibrant community of researchers exploring diverse approaches to parallel programming — languages, libraries, compilers, tools — and their applicability to multiple application domains.

Microsoft researchers are investigating all of these approaches, from coordination languages for robots and distributed systems to mobile phones to desktops and data center clouds. To engage the academic community, Microsoft funds multicore research projects and many sites, and we have partnered with Intel to fund the Universal Parallel Computing Research Centers (UPCRCs) at the University of California at Berkeley and the University of Illinois at Urbana-Champaign.

As Richard Hamming famously noted, “The purpose of computing is insight, not numbers.” In that spirit, I believe our research challenge is to break free from the limitations of the desktop metaphor and exploit the ever greater performance of multicore chips to create new human-computer interaction metaphors that are more natural and intuitive. This will require new approaches to parallel computing education and increased collaboration with researchers in application domains.

As an example, consider one possible future — “spatial computing” — where real-time vision and speech processing, coupled with knowledge bases, distributed sensors and responsive objects, enhance human activities in contextually relevant ways while remaining otherwise unobtrusive. Such an infosphere would adapt to its user’s needs and behavior and move seamlessly across home, work and play.

Multicore brings enormously interesting intellectual challenges and the opportunity to rethink much of how we approach computing.  Let’s embrace the opportunity!

– Daniel Reed is Microsoft’s Scalable and Multicore Computing Strategist and a member of the President’s Council of Advisors on Science and Technology (PCAST). Contact him at reed@microsoft.com or his blog at www.hpcdan.org

–

Multicore (parallelism) represents a fundamental challenge and change for all of computing and computer science. It represents the fundamental constraints of physics — nature loves parallelism — surfacing and interacting with some fundamental tenets of computing. We have formulated our theory of computation and complexity primarily on sequence — in control and state. Fundamental physics (and consequently circuits and architecture) which makes parallelism fundamentally cheaper is now challenging us to broaden the foundation of computing with parallelism as a first class element. I believe that as a research community, this is a first-order challenge to respond — in nearly all disciplines of computer science. First and foremost, this is a major intellectual challenge to the computer science community to “reinvent” or at least broaden computer science in this way. Second, this is a major educational challenge, where students we are training today to think about “Computer Science founded on sequence” are being launched into a world of parallelism. For their benefit, we must mount a rapid response in pedagogy and curriculum to ensure these students emerge armed to deal with the future of computing in their careers.

Now, let me turn to research funding in parallelism — which is a critical need in all areas from architecture, runtimes, compilers, programming languages, algorithms, and theory. We need major increases in funding and research activity in all of these areas. Governments must take the primary role in funding research in information technology for the long term economic development and societal well-being. We would like to see aggressive large-scale funding of long-range research in parallelism, and that the fruits of that research be made broadly available for commercialization. In the U.S., DARPA has a long track record of funding such research in IT, but such investment has decreased in recent years. We would like to see it increase both in DARPA, as well as other parts of the US government, and yes around the world. History has proven that only governments are able to invest in this type long-term general economic development, and it is critical that the research outputs be generally available for society at large to benefit — not just a small population of gatekeepers. It is great to see this vision being pursued in many regions around the world.

At Intel, we depend heavily on a broad range of science and engineering research pursued by the global academic community. Many of the innovations we commercialize were first conceived in universities — often many years before their practicality — and we have contributed additional innovations and refinements to bring them to the broadest swath of society possible. We strongly support (and contribute our time, money, and leadership to) the health of the research and innovation community globally. We have made significant investments in education (multicore curriculum and training) and research (research grants, Universal Parallel Computing Research Center with Microsoft) for parallelism, and continue to encourage others to join us in doing so.

– Andrew Chien

–

To see the first article in this series, click here.

As I mention on my personal blog (at http://csdiary.org), this points out a somewhat strange aspect of computing research, namely that there isn’t much computing research in the major core-science publications. I’m thinking specifically of Science magazine, Nature, and PNAS. In fact, I took a quick scan over the past 5 issues of Science and Nature.  Over those issues, in Science one sees 35 research articles and reports in the biology and medical science areas, 14 in chemistry/materials, 10 in earth and atmosphereic sciences, 5 in astronomy and astrophysics, and several in physics, psychology, and archeology. Only one article in computer science!

In Nature, the situation is even more stark. In the last 5 issues we see 11 research articles in biology, 2 in chemistry, 1 in astrophysics, and 1 in psychology. None in computer science.

Why should we care about this? Well, lately the computing research community has become very concerned about its “image”, particularly in the lay public (including, notably, the US Congress). Yes, we want people to know the full impact of computing, the range of jobs and activities the computing professionals are involved in, and the great economic benefits the come from our research. But we also need, in the interests of public education and our image, to explain computing research to the world’s science scholars. Doing so not only puts our research to a good test, but it also helps to cast an aura of intellectual respectability that would undoubtedly contribute positively to the image of the field.

There could be important consequences within the federal government, too. I asked Peter Harsha, the director of government affairs for the CRA, what he thought about this. Here is what he said:

I think all three [Science, Nature, PNAS] generate news in the more mainstream press that gets noticed by Members of Congress and Administration folks. So while most policymakers and their staff generally don’t read the periodicals directly, the noteworthy stuff they publish finds its way into the NY Times, WSJ, or Washington Post, which quickly gets policymaker attention.

I think all three publications have a good track record of generating that buzz in the mainstream press (Science and Nature, especially).

As we as a community work on getting our government to step up its support of basic science research, to what extent will our representatives include computer science and engineering? While computing will be hard to forget in any serious discussion about funding priorities, putting ourselves “front and center” in these sorts of publications should help not only the cause of computing research but also the large cause of scientific research.

– Peter Lee