Slashdot Mirror
Teragrid: Massive Grid Computing
onyxcide writes: "Envision is running a quick article on a new national grid of computing resources called TeraGrid. Half a petabyte of disk storage, 40-gigabyte-per-second national optical backbone, and 13 teraflops of computing power will make up this monster. It will allow "lavish amounts of online data to be continually available for instantaneous analysis, data mining, and knowlege synthesis." There's another article in the same magazine here: Transforming Research with High-Performance Grid Computing" LighthouseJ adds some details: "C|Net's news.com has a story about a new Compaq supercomputer named Terascale. It uses 3,000 Alpha EV68 processors distributed over 750 servers using networking systems from Quadrics. They say it can perform as fast as 10,000 desktop PC's combined in one second. The massive computer will make it's official debut on Monday at the Supercomputing Center in Pittsburgh PA."
Wow, some part of my mind is drooling!
;)
The groundwork for a Matrix/Johnny Mnemonic-style cyberspace, anyone?
I tried combining 10.000 PC's once. Took me, my friends and a bulldozer all weekend.
If you're going to be in Denver the week of Nov 12, 2001, consider stopping by. If nothing else, the place will have free and open 802.11b!
-- Stanislav Shalunov
... it is the social (read human political) barriers. How often do you get organisations which have a disjoint culture agreeing to "share" resources? Scientists are no different as there's only a limited pot of government funds ... ask the Department of *ENERGY* why they're doing genome research. Corporate cultures makes it difficult to merge, just putting together a backbone and lots of honey pots (CPU resources) doesn't automatically lead to a automagic collaboration of trust. Grid computing doesn't address the social issues ... will my work be safe?, can I get a fair cut of the machine, will commercialisation contaminate standards, etc ...
... someone in a distant /. post noted that the GPL promoted a wierd form of trust ... because you knew the viral nature would eventually force publishing of any improvements, you had some confidence that the effort you put into developing software would (potentially) be amplified giving you improved down the track. The Sun Community Source License (SCSL) and Microsoft End-User License (MSFU) don't exactly inspire the same confidence and level of trust.
... but nothing different from a fancy queuing system. Other systems such as Globus are seriously researched but writing apps is still difficult. As for Microsofts .BET, it is stilll an unknown factor (and RPCs over low-latency internet doesn't exactly promote radically new killer apps). What does it require for a radically new level of trust (integrity, availability, confidentiality) to engineer the new killer apps? Chucking money at hardware without solving the human issues seem a little like an indirect government subsidy to the chip companies to me.
My point is that it takes a while for *HUMAN* systems to adjust to new technology waves. I would point out that in the early 1900s, factories were driven by belt-pulleys and machines (lathes/drills/press/etc) were contained in small 3-story buildings. Once electric motors got small enough and eliminated the physical requirement of being mechanically linked to the power source, then we could suddenly build whole acres of assembly plants and skyscrapers.
I see a necessary transition for software
Currently TeraGrids are the beowulf of ASPs
LL
Sorry, I had to say it. Actually, XP is kinda purty. Been playing with it the last couple days and haven't gotten it to crash... yet.
Well, the parachute pants might slow him down a bit.
Got to hear a talk from Henri Casanova, one of the top dogs working on distributed application scheduleing and simulation software for The Grid. Neat stuff, but, as he addressed in his talk, we're really looking at a network of computers that only people needing massively intensive computations done on highly parallizable problems would find useful. Translation: only researchers in certain fields need this.
I could not justify my existence if I were a turkey farmer. Would I terminate myself? Undoubtably, yes.
FYI, this will be updated after supercomputing, but this is the list of the fastest computers in the world.
The Pittsburgh's super'puter will rank up there with LANL's new one (also a Compaq based one). Pittsburgh's will be the fastest SC for nonclassified work.
I'm not sure whether or not it'll dethrone LLNL's ASCI White or not. It does knock seaborg @ NERSC from the fastest unclassified SC spot though.
Do you know why the road less traveled by is littered with the bones of the unwary?
Ahhh, I can't touch that comment.....
Moore's law refers to the number of transistors that will be packed onto an integrated circuit. Doesn't say anything about processing power let alone massive computer arrays.
"Prefiero morir de pie que vivir siempre arrodillado!"
Hmm... think long and hard about that one! I'm not even stoned, and its messing with my mind.
"Knowledge Synthesis"... doesn't that defeat the point of "knowledge"? whoa.
>/dev/null
Skiers and Riders -- http://www.snowjournal.com
There would be plenty of room for speculation, and participants in the market would basically be betting on Moore's law, in addition to the other economic factors common to all derivatives markets.
The problems I forsee are to do with the standardization of the contracts. We would need to agree on an architecture, and a delivery method for the CPU cycles. All in all though, this could be a really lucrative business, especially with the demand for GHz from Hollywood movie studios set to explode in the near future due to actors being replaced with CGI animation.
Sometimes I feel like I am living in a Bruce Sterling or William Gibson novel, the pace of technology just seems to get faster and faster.
All please read that announcement as having said "40 gigabit," 'cause that's what it is. Still fast... 4x OC-192.
God knows how the research people pay for this. Impoverished corporations like my employer still dick around with multiples of T1.
Avaki were in peddling their grid computing solution, and I had to say to the guy... "do you have any idea how little bandwidth we have?"
Grid computing will affect the rest of us when everyone can get high speed network connections.
I have a friend who's currently trying to think of a business case for Grid Computing ... but is having trouble. Apart from academics and researchers, can The Grid ever become mainstream? Why should companies invest in it, i.e., your average medium-to-large corporation? The books often seem to cite creation of virtual companies and vertical integration of companies (i.e. from the component manufacturers to the end retailers), but these situations don't seem particularly realistic ... and you'd have to agree policies over data sharing for a start!
I find this area of research particularly interesting because of my own research and the high amounts of computing power that it requires.
But to answer some of the previous posts about the sharing of resources, one of the larger problems is to figure out and method of saying this:
Run program X at site Y under policy P providing access to Z under policy Q.
So, it's not like you'll just be able to tap in, there will be policies for program execution and data access. But it's coming faster than you think.
One of the coolest concepts is that of process migration which will probably be integrated into a ubiquitous computing grid. Whereby a process running on Processor A, Architecture X can migrate to Processor B, Architecture Y and preserve state. I've seen this work with some DEC's and Sparcs swapping processes and it's most impressive, but still needs some work.
I would suggest reading The Grid: Blueprint For a New Computing Infrastructure if you'd like to get more about the general idea of the grid. It's light on technical details, but a good high point view.
My Slashdot account is old enough to drink...
Keep in mind that only certain applications would work on a massively distributed basis. Things like Seti@home and Distributed.net are good because they deal with small chunks of data that can be processed by PCs. Things like CG rendering probably wouldn't work in a broad sense because of the kinds of bandwidth and storage needed to deal with frames. I have no idea how big (in bytes) a single frame of a motion picture is, but I would guess that the costs in bandwidth just to send back the finished product would neglect any benefit.
:-)
At the same time, I'm working on some artificial intelligence research, and I could definitely benefit from having computers spread around doing my work. I'd probably even pay for it.
A speech...