| |
Search videos for computation |
|
|
|
|
Disk-Based Parallel Computation, Rubik's Cube, and Checkpointing
Google Tech Talks
March, 24 2008
ABSTRACT
This talk takes us on a journey through three varied, but interconnected
topics. First, our research lab has engaged in a series of disk-based
computations extending over five years. Disks have traditionally
been used for filesystems, for virtual memory, and for databases.
Disk-based computation opens up an important fourth use: an abstraction
for multiple disks that allows parallel programs to treat them in a
manner similar to RAM. The key observation is that 50 disks have
approximately the same parallel bandwidth as a _single_ RAM subsystem.
This leaves latency as the primary concern. A second key is the use
of techniques like delayed duplicate detection to avoid latency. For
example, hash accesses accesses can be saved (even saved on disk), until
there are sufficiently many pending accesses to use standard streaming
techniques. We have designed a library for search problems that exploits
the high parallel bandwidth while hiding the latency. We build
abstractions for search that employ parallel disk-based hash arrays
with the same speed as a single hash array in a single RAM subsystem.
In the case of Rubik's cube, we exploited this mechanism by using
seven terabytes of distributed disk in a search problem that showed
that 26 moves suffice to solve Rubik's cube. Our initial efforts
emphasize idempotent operations, so that we can easily recover from
hardware or software faults. We next intend to apply a more general
solution for fault recovery: checkpointing. This separate effort
in our lab has now produced a mature, robust user-level checkpointing
program has now matured. The package works successfully in tests
on OpenMPI, MPICH-2, OpenMP, and parallel iPython (used in SciPy and
NumPy). Our DMTCP package transparently checkpoints parallel,
multi-threaded processes, with no modification either to the
operating system or to the application binaries. Extrapolating
from current experiments, we estimate that we can checkpoint a 1,000
node parallel computation in a matter of minutes. We are currently
searching for a testbed on which to demonstrate this scalability.
Speaker: Gene Cooperman
Length: 4433
Rating: 4.30 (17 ratings)
Tags: google techtalks techtalk engedu talk talks googletechtalks education
|
Play |
|
|
Nereus: Massively Parallel Computation in Java
Google Tech Talks
May, 8 2008
ABSTRACT
With 1 Billion idle PC's in the world, there is approximately $100 Billion's worth of CPU time going to waste every year. Given that insatiable global demand for computation is fuelling massive datacentre proliferation, with consequential environmental impact, surely there must be a way to leverage idle time on desktops to relieve some of this pressure?
In the past, the term "Grid Computing" has been used to encompass the idea of using idle desktop capacity productively. Unfortunately "Grid Technology" has yet to provide a solution for this global problem, especially for the vast majority of machines which are outside the corporate firewall. Nevertheless, there is a class of problems which are ideally suited for idle desktops and can leverage millions of machines if managed effectively. We use the phrase "Massively Parallel Computation" to describe such tasks, by their nature beyond the scale of any datacentre, but also beyond the capabilities of virtually all grid software too!
Nereus is a new, open source, pure Java MPC platform which finally offers an easy way to use the world's idle computers. Joining a computational project running on Nereus is as simple as clicking a link in a web page, and it provides a secure sandbox to protect the host computer from malicious software -- just like the Applet sandbox in a browser. Developing a Nereus application is also simple; it's virtually identical to building a servlet for a J2EE container, and the progamming concepts will be familiar to all Java developers (J2EE focussed or not). Nereus applications are as simple to write as any Java applet; one JAR on the classpath, one class to extend, and a small but powerful API.
Perhaps most importantly, Nereus provides a simple way for owners to delegate control of their machine to resellers, who are then able to group machines together and offer attractive resources to customers. This feature also provides for the owner getting paid for this 3rd party use -- something drastically different from many MPC projects which require people to donate their spare computer time (e.g. SETI@Home).
This talk will describe Nereus, its architecture and implementation details, as well as a step-by-step demonstration of building a "Hello Nereus" MPC application. Other demos include showing a MPC 3D rendering application -- animating a movie lightning fast. The audience will also be invited to join a Nereus MPC application created and started within the talk, using their laptops from where they sit. We can then all experience the CPU power unleashed by simply sitting in a talk!
Nereus is simple x 3; simple to deploy, simple to join, simple to use. By demonstrating this during the talk, and by showing the successful work being done at Oxford University using it we hope to simulate interest in the more widespread use of this open source technology in the next generation web.
Oh, and what about using "native" code on Nereus resources? We will show how, still within the full Java security sandbox, by using the JPC pure java x86 emulator (see JavaOne 2007/2008 and http://www-jpc.physics.ox.ac.uk )
Further information about Nereus can be found at http://www-nereus.physics.ox.ac.uk
Speaker: Dr Rhys Newman
Chief Architect IGD Group, Dept Physics Oxford University (JavaOne "Rock Star 2007")
Rhys joined the IGD group in 2004 after spending over 20 years as a programmer in industry and academia. The first 10 were in C/C++ but since then he has focused exclusively on Java technology. Rhys got his DPhil from Oxford University in 1998 and since then has worked in several technology start-ups and academic positions.
Speaker: Dr Jeff Tseng
Group Lead, Grid Technology, Dept Physics Oxford University.
Jeff formed the IGD group in 2003 when he arrived at Oxford to take up his post as University Lecturer in Physics. Before that, he was a research scientist at MIT and Fermilab working on the CDF particle physics experiment, where he built data acquisition systems out of off-the-shelf PC's and was in charge of the experiment's data handling systems and initial forays into grid com...
Length: 1651
Rating: 4.10 (8 ratings)
Tags: google techtalks techtalk engedu talk talks googletechtalks education
|
Play |
|
|
Human Computation
Google TechTalks
July 26, 2006
Luis von Ahn is an assistant professor in the Computer Science Department at Carnegie Mellon University, where he also received his Ph.D. in 2005. Previously, Luis obtained a B.S. in mathematics from Duke University in 2000. He is the recipient of a Microsoft Research Fellowship. ABSTRACT
Tasks like image recognition are trivial for humans, but continue to challenge even the most sophisticated computer programs. This talk introduces a paradigm for utilizing human processing power to solve problems that computers cannot yet solve. Traditional approaches to solving such problems focus on improving software. I advocate a novel approach: constructively channel human...
Length: 3091
Rating: 4.80 (25 ratings)
Tags: google human computation esp game captcha
|
Play |
|
|
Efficient, Secrecy-Preserving, Provably Correct Computation (and Some Cool Ap...
Google Tech Talks
July, 21 2008
ABSTRACT
Cryptography is now much more than keeping credit card numbers safe from packet sniffers and laptop thieves. We combine several advances in cryptography theory to construct a model of computation in which a third party securely performs computations for a set of parties who do not trust one another. These parties encrypt their inputs to a computation, then circulate the encrypted values. Our third party decrypts them, performs the computation and then issues correctness proofs of the results: we cryptographically "tie its hands" to do the right thing. In addition to this model of provably correct computation, we will discuss other techniques to control the information flow of secret data to and from the party, so even the third party cannot profitably abuse the secret data before or after it knows it during the computation. Finally, we will illustrate the power of these techniques in e-commerce, and our design decisions, through secure electronic auctions and securities exchanges.
Speaker: Dc. Christopher Thorpe
Dr. Christopher Thorpe is a computer scientist and entrepreneur who recently completed a Ph.D. in computer science from Harvard, advised by Michael Rabin and David Parkes. Prior to obtaining his Ph.D., he led internationalization at Tellme Networks, after being the internationalization engineer for Yahoo!'s commerce properties and software engineer for Yahoo! Merchant Solutions. Chris joined Yahoo in 1998 when it acquired Viaweb as its Yahoo! Store property; Viaweb was founded by graduate students he knew as an undergraduate at Harvard. He also holds an A.B. from Harvard in Computer Science and Music.
Length: 4285
Rating: 4.90 (11 ratings)
Tags: google techtalks techtalk engedu talk talks googletechtalks education
|
Play |
|
|
Human Computation
Google TechTalks
July 26, 2006
Luis von Ahn is an assistant professor in the Computer Science Department at Carnegie Mellon University, where he also received his Ph.D. in 2005. Previously, Luis obtained a B.S. in mathematics from Duke University in 2000. He is the recipient of a Microsoft Research Fellowship. ABSTRACT
Tasks like image recognition are trivial for humans, but continue to challenge even the most sophisticated computer programs. This talk introduces a paradigm for utilizing human processing power to solve problems that computers cannot yet solve. Traditional approaches to solving such problems focus on improving software. I advocate a novel approach: constructively channel human...
Length: 3094
Rating: 4.80 (9 ratings)
Tags: google human computation peekaboom esp game
|
Play |
|
|
Quantum Computing Day 1: Introduction to Quantum Computing
Google Tech Talks
December, 6 2007
ABSTRACT
This tech talk series explores the enormous opportunities afforded by the emerging field of quantum computing. The exploitation of quantum phenomena not only offers tremendous speed-ups for important algorithms but may also prove key to achieving genuine synthetic intelligence. We argue that understanding higher brain function requires references to quantum mechanics as well. These talks look at the topic of quantum computing from mathematical, engineering and neurobiological perspectives, and we attempt to present the material so that the base concepts can be understood by listeners with no background in quantum physics.
This first talk of the series introduces the basic concepts of quantum computing. We start by looking at the difference in describing a classical and a quantum mechanical system. The talk discusses the Turing machine in quantum mechanical terms and introduces the notion of a qubit. We study the gate model of quantum computing and look at the famous quantum algorithms of Deutsch, Grover and Shor. Finally we talk about decoherence and how it destroys superposition states which is the main obstacle to building large scale quantum computers. We clarify widely held misconceptions about decoherence and explain that environmental interaction tends to choose a basis in state space in which the system decoheres while leaving coherences in other coordinate systems intact.
Speaker: Hartmut Neven
Length: 3387
Rating: 4.80 (63 ratings)
Tags: google techtalks techtalk engedu talk talks googletechtalks education
|
Play |
|
|
Graphics Processors, Graphics APIs, and Computation on GPUs
Google Tech Talks
December 5, 2006
ABSTRACT
Graphics processing units (GPUs), designed to produce images ranging from scenes in computer games to CAT scan results, have become computational powerhouses capable of hundreds of gigaflops per second. As computational power has increased, so has flexibility and programmability, allowing GPUs to be applied in new ways in both graphics and non-graphics applications.
I will review the architecture of recent GPUs and the traditional software interfaces to them, as well as present a new interface from ATI that more closely reflects current GPU design and use. Credits: Speaker:Mark Segal
Length: 3645
Rating: 5.00 (1 ratings)
Tags: google howto graphics processors apis
|
Play |
|
|
Computation & Journalism Symposium
What is computated journalism? And how is it changing the journalism profession? Interview here with Michael Skoler, Executive Director, Center For Innovation In Journalism, American Public Media. Taped during the Computation & Journalism Symposium, Georgia Institute of Technology, Atlanta, Georgia, February 23, 2008.
Length: 449
Rating: 4.00 (3 ratings)
Tags: computated journalism symposium Atlanta Georgia Tech Center for Innovation Michael Skoler American Public Media
|
Play |
|
|
Good slur computation in Finale 2007
Makemusic Finale has definitely a better slur computation algorithm than Sibelius 5...
Length: 39
Rating: 3.70 (3 ratings)
Tags: Finale Sibelius notation software bug
|
Play |
|
|
Matlab Distributed Computation
demonstration of matlab distributed computing tool box.
Length: 543
Rating: 0.00 (0 ratings)
Tags: Matlab Distributed Computation
|
Play |
|
|