Slashdot Mirror
Canada to Launch Countrywide Virtual SuperComputer
LadyCatra writes "A serious shortage of world-class computing power in Canada prompted University of Alberta scientists to create the next best thing -- a countrywide, virtual supercomputer.
On Nov. 4, thousands of computers from research centres across the country will be strung together by a U of A effort to create the most powerful computer in this country.
The full story is here"
Actually, they are Linux Clusters.
I was visiting the Vancouver site a couple of months ago when they were assembling it. It looks sweet. A nice big array of Dual Athalons. The system is being linked together over CA*Net 3, a nation wide OC192 fibre network.
They're also experimenting with distributing different parts of the system in different locales. Like disk storage in one part of the country, heavy number crunchers in the other, to see how distributed a system can really be and still function well.
CA*Net is still looking for applications, the network is being severely underutilized. http://www.canarie.ca/advnet/canet3.html
The article isn't very specific on the kind of problems they will try to solve. The 'search' problems, where you have a big search space than can be easily divided into smaller chunks are easy. Unfortunately some problems cannot be easily split into many independent parts - simulations generally fall into this category. Weather simulations, nuclear explosion simulations, well, simulations in general :-). You can just assign each computer a square mile of terrain, do the computations for the whole simulations, then merge the results - the neighboring squares interact, so computers have to communicate after each time slice. This is where communication will probably slow your 'network supercomputer' down. No matter how fat the pipes are, they will be several orders of magnitude slower than an internal supercomputer bus in terms of latency. To put it short: this might be of some use, but they better start gathering money for a real supercomputer.
And what do you call this?
The computers will be linked by the Internet, but involve a simple networking system, Lu said. Keeping the linkage as simple as possible was the goal.
Read the article the next time, will you?
Civilization is the process of setting man free from men.
From the article Gerald Oakham and his fellow physicists have a problem. In the hunt for the most elusive speck of matter known to science, they are about to generate more data than any computer on the planet can analyse.
You mean like Popular Power tried (and failed)to do? Check their old site to see what they used to propose.
Looks like selling CPU cycles is not a lucrative business...
I code, therefore I am.
Google is doing this. Click on a button in the Google Toolbar, and your compute starts number crunching in its idle time.
Check out the Google Compute Faq and the Kuro5hin discussion on the subject.
I think what this really needs is to be make easier for the mainstream, so anyone could do it. Perhaps bundle the tools (programming and deployment) with mainstream operating systems?
Sun have Grid Engine and I believe Intel have something similar. The issue is that this kind of distributed processing is only useful for problems that can be divided into many discrete subtasks, which do not need to interact with other nodes while they are running, otherwise the work you need to do to communicate between nodes slaughters performance (that's why clustering hasn't taken over the world, vertical scaling on an active backplane is still the best solution for most jobs). The typical corporate large-compute job is data mining or decision support, neither of which scale particularly well horizontally.
Yeah. There's actually quite a lot of research going into this currently. It's called the Grid (think "power grid", ubiquitous, simple to use), and I predict it will be the next big buzzword.
See Global Grid Forum, Grid Today and the Globus project for starters.
The problem of buying and selling computation power on some sort of broker basis is a quite interesting problem in itself. Exactly what are you selling? Hardly CPU hours, since the value of those depends on the hardware.
Looks similar to the Grid Computing project from India, announced sometime back ...
I would imagine this distributed computing effort will be conducted using Ca*NET, which isn't really what I'd consider a public network. As such, I don't think they'd really have to worry about hackers and such.
The 'Canadian Internetworked Scientific Supercomputer (CISS)' website is located here: http://www.c3.ca/ce/ciss_t.html
It seems that November 4th they will be doing a full 'production' test. Cool.
-- bartman
Heh heh, the U of Alberta hosts the web and ftp space for OpenSSH and OpenBSD.Also, Bob Beck works at U of A. Bob helped develop the first OpenSSH release, not sure how active he is these days.
For U of A, that all adds up to "premium class" tech support for anything to do with SSH.
Virtual Laboratory of Eastern Ontario.
The High Performance Computing Virtual Laboratory (HPCVL) was formed by a consortium of four universities located in Eastern Ontario (Carleton University, Queen's University, The Royal Military College of Canada, and the University of Ottawa).
http://www.hpcvl.org/
It's also in the Top 500 supersomputer list, so it must be half-decent. So if four universities can have a dencent computer in Canada, others probably do too.
I believe they will use high speed networks of Linux based Beowulf clusters (actually clusters of clusters of clusters). Ontario has already established SHARCNET between a number of Universities with a total of over 500 COMPAQ Alphas (mostly four-processor, 833Mhz, Alpha SMPs) and some Pentiums, all running Linux. A press release from last year gives a good overview of the project, already first in Canada and the 11th most powerful academic computing system in North America. I believe the Canada wide project will essentially form a cluster of these cluster of clusters.
SHARCNET has been up and running for a while and last year accounted for about 27% of supercomputing power in Canada (half of all supercomputing power in Canadian universities), with three sites on the Top 500 list and total power exceeding institutions like Cambridge, Princeton, Cornell and Caltech. There's loads of information available about the hardware and software used at each facility, as well as CPU load and usage statistics at members sites like these status charts from the most powerful individual site, at the University of Western Ontario. As for applications, a number of researchers are already using the system for a variety of projects across science, engineering, and economics.
My next sig will be ready soon, but friends can beat the rush!
Thanks for the interest in our project. We have composed a http://www.cs.ualberta.ca/~ciss/CISS/faq.html. I hope it is useful.