Slashdot Mirror
Supercomputer To Use Optical Router
Izmunuti writes "From a NYTimes article: 'Highlighting a radical departure in the design of the fastest computers, the California Institute for Telecommunications and Information Technology plans to announce on Monday that it will use an optical router
designed by a Texas company as the heart of a
campus-wide supercomputer that will be woven together with optical fibers.'"
At first we wanted everything to be transfered from analog type formats to pure digital, now we are exploiting analog formats (like optics) to get insane speeds.
Will we go back to all digital in about 10 years?
So how is this any different from using gigabit fiber instead of copper? Although for most large clusters gigabit is too slow when you need to move around a terabyte of data. Look into NUMA.
Only the State obtains its revenue by coercion. - Murray Rothbard
Someone figured out that you can pack more bandwidth and less latency into fiberoptics than copper?
More importantly they are actualy using an optical router to prevent what has become a botleneck in resent years. I.e. Data comming off a fiber pipe is converted to electrical signals before being routed to it's next destination where it's converted back to little bity laser beams.
This should be faster than your typical loadsharing super computer (SETI@home) but slower than the miranet using hardcore. With enogh nodes however there is no telling howfast this baby can get.
--= Isn't it surprising how badly I spell ?
When it comes down to it, the computers do the work. You can do useful supercomputing with almost no networking, you can't do useful supercomputing with blindingly fast networks and no computers.
(Somehow, the quote reminds me of people who think that managers and lawyers are the important part of a company, and engineers and customer service are a nuisance to be minimized.)
Anyway, we're about pushing the limits of copper, with 1000bT, and I'd imagine network speeds will only continue to climb with increased use of fiber. I can see, in 5 to 10 years, optical switches becoming more common in office environments as file sizes and network speeds continue increasing.
From the article: "We're moving to an optical-centric world in which the computers are the slow things and you reluctantly add them in," Dr. Smarr said.
Umm yeah... If you didn't add in the computers what good would the optical network be?
I think this is a great example of what the old saying "think before you speak" is meant to help one avoid.
RTFA... " Each of the clusters is based on Intel microprocessors and runs the Linux operating system."
I'm not seeing how it's all that revolutionary. Am I wrong in saying it's essentially a Beowulf connected by an optic network?
Do we need to sign up for these accounts, or is there a free way in?
Do they track IP's so could we make a universal slashdot login?
A pure optical router using analog signals which are passed through a crystal and output at certain locations based purely on their wavelength(wich coresponds to the exact binary data of the full packet) and the path which the light beam is forced to take! HA! Ha HA! MWUAHAHAHAHAHAHAA!
>:D
What? It could happen...
-- 'The' Lord and Master Bitman On High, Master Of All
guess what.. its *almost* faster than the speed of light. well, if it wasnt light of course :)
Optical Router Uses Supercomputer
Also, their statement on the Chiaro Networks "OptIPuter" is here. Caltech is an entirely different animal.
"Prepare for the worst - hope for the best."
Just as the article started to pique my interests, it was over. That sucks!
Yet another technology article written without any real information. I realize in writing you are supposed to write to the common reader, but sometimes it seems like they would be better off not writing about it at all if they didn't intend on clueing us in on any of the facts.
A slip of the foot you may soon recover, but a slip of the tongue you may never get over. -Benjamin Franklin
The optiputer will initially consist of about 500 processors linked via the optical switching system that will permit parts of the computer to share information at the speed of light.
In other breaking news electromagnetic radiation (read: electricty) doesn't travel at the speed of light! Coming Soon to Fox: When Reporters Get Confused
At any rate that article was darn short on details, and the company's website wasn't any better. Anybody have any relevant data on exactly how fast this switching system is? I'm curious about their optical router at the heart of the system as well. It is my understanding that the slowest part of any fiber-based system is the router since the signals must be converted from light to electrical than back to light signals again. One would assume that such a design would be entirely too slow to be used as a bus. Of course, I may be entirely wrong...
Good God, is there anything that Dirk Pitt can't do?
;)
NUMA must be -really- branching out.
I'd like to see a beowulf...
Oh, nevermind.
Electrical signals travel at around 1/10 the speed of light.
Yay for free subscriptions.. here are some other sources for similar reportings that don't require evil subscriptions.
"Somehow, the quote reminds me of people who think that managers and lawyers are the important part of a company, and engineers and customer service are a nuisance to be minimized."
You mean like Microsoft?
Or most ISPs?
In the free world the media isn't government run; the government is media run.
This would be a network of computers, not a supercomputer. The definition is becoming more lenient, so in a few years everyone on the internet will be a node in the world's largest (and only) supercomputer, and 80% of them will be redundantly running through Windows DLLs. Yay.
The idea of optical routing is that, even in typical gigabit or any optical based networking media, the bottleneck is the processors in the routers. This is because the light must be converted to electrical signals, and then routing decisions and switching are done on the processor of the router. After being processed, the signals are converted back to optical to be sent out the appropriate port.
Optical switching means that the light coming in on fiber from different devices is never converted to electrical to be routed. The actual light signals are switched from port to port. This was originally planned to be done with very small mirrors! (no joke!) which would aim incoming light to the corresponding outgoing port.
According to the whitepaper on Chiaro's website, they have found a way to avoid the mirrors (which have an obvious bottleneck themselves, as well as potential mechanical failure) and they are able to multiplex or switch the light based on applying an electrical field to some of the optical components which them changes the angle and therefore the destination of the light.
Most of the key people at Chiaro are people who jumped ship from Convex Computer after they were acquired by Hewlett-Packard back in the mid-90s. Convex's claim to fame was to have invented and productized the first mini-supercomputer hitting the sweet-spot between the biggest vax and the smallest cray and they were very successful for about a decade.
Larry Smarr, of UIUC's supercomputing center (aka the place where Mosaic was developed) has always been a big fan of the Convex crowd.
Another bit of trivia - Jeff Christenson, of PERL fame is a convex alum as well as Dave Taylor of Id Software fame and a whole host of other key people now scattered about the world.
When information is power, privacy is freedom.
HAL will be born a few years late...
Strange women lying in ponds distributing swords is no basis for a system of government.
Interesting. How is he a racist troll by linking to a page that promotes no special treatment based on skin color? Equal Opportunity is reverse discrimination. Get out of the leftist, bleeding heart arts colleges..
A new account has been created for the benefit of slashdot users who don't care to register with NYTimes.
Username : SDUser
Password : slashdot
enjoy everybody
click here to login.
>"the communications lines will be the fastest part of the computer and the processors will become slower "peripherals."
Just imagine....
---
Open Source Shirts
...And it still gets Slashdotted...
The point is that a Californian organisation is using something "designed by a Texas company". Now that is "a radical departure"!
:)
Maybe I'm reading this wrong?
OLPC Australia
Why do I get the feeling that any day now a liquid metal asassin is going to appear and start taking people out? ...heading to my bunker
Corect me if I'm wrong but I always thought SGI was using light in it's interconnectors between machines? That's how they can achieve amazing throughput.
-- Leeeter than leet
There is an old article in Science News that demstrates things being sent faster than light. The interesting bits of the article talk about one experiment where a beam of light was shot into a cloud of gas, and it exited the cloud before it entered it. Another has them sending Mozart 40th symphony at 4.7 times the speed of light.
And btw, 6.6 microseconds aint bad. I never read below, but I can imagine the massive amount of (bad) Beowulf jokes. I doubt their latency is any better. Assuming only 1 packet is sent at a time, thats ~150,000 packets per second (theorectical peak of course). Seeing as how I havent sent that many in the last 9 hours this isnt too bad of a problem. You are right that this isnt entirely whats needed for this purpose, but you cant learn if you dont try.
http://www.jacobsschool.ucsd.edu/news_events/news_ 2002/20021118.shtml
Amazing.
I was told by a fiber-optic transceiver engineer that signals actually travel faster in copper coax than in fiber (in both it's less than c, the speed of light in a vacuum). So couldn't you get even better results by hardware-switching a coax signal? And how usefull is only being able to talk to 1 other node at a time? Sounds to me like these guys have reinvented the T-bar used to connect IMB System 370 channels together... (albeit with much better performance).
"Freedom means freedom for everybody" -- Dick Cheney
Because the previous attempt to implement a campus wide supercomputer using an RFC 1149 based network caused the pigeons to burst into flames.
i am the cl-itit commander. no one rules the cl-itit like me. not this little fucker, not any of you little fuckers...
racist troll? Click here to see why the parent post is tacubs. If you don't know what tacubs means, please feel free to look at my journal.
Support Israeli punk bands. Man Alive.
There is still some room to grow left in the semiconductor business....but not much. Now that chip makers are near the limit in the "how small can I shrink it race", the goal now is to fit more die on single silicon wafer. As the "need for speed becomes more critical I think optical computing breakthroughs will become more common place as soon as the semiconductor business and research community start to move away from silicon semiconductors. When the focus on optic technology becomes dominant silicon chips won't go away, but they won't be as high tech.
"You helped our nation celebrate its bicentennial in 17 -- 1976." --George W. Bush, to Queen Elizabeth, Wash
The experiment you reference does NOT show information travelling faster than light.. as is explained in the article.
The waveform appears to exit the apparatus before it enters, but this is not so under scrutiny... as the article says, the beginning of the wave enters the glass (long before the peak) and there is enough information there to re-create the original wave.
There are several phenomenon that appear at first to be superluminal, but they do not violate relativity, and are not actually moving anything faster than light, nor are they transmitting information.
look, fiber is awsome. it is not prone to electrical disruption like other digital media. plus, the speeds are very nice.
-illumina+us "I put on my robe and wizard hat..."
Imagine a Beowulf cluster of these things!
Most 0 mods I have ever seen in a /. article. This must tell us something...
Here comes another one.
He still drives a Porche with Convex as number plate.
Help fight continental drift.
Sun does this too
The pure optical IP or ATM router is still years away. Optical computing isn't up to optical packet decode and route lookup. Optical buffering isn't ready, either, although you could potentially store packets temporarily in a fibre delay line.
While company calls it optical phased array, "array" only works with linearly increasing electric field, what turns it into just another case of diffraction, with "waveguides" being more like segments of diffracting material changing their properties each as a whole when electric field is applied (as opposed to pressure being applied in other devices). It would be more impressive if the device was purely optical -- if some material changed its properties based on the light applied to it, and bent another beam because of that change.
Contrary to the popular belief, there indeed is no God.
I mean, you pour a six pack on a computer, it just won't go as fast. OK, so you might need some work to beat the fault tolerance, but I'm sure you can find six parts that'll bring it down together.
with a wavelength dependant no linearity in the medium would do exactly that.
An FP inferometer consits of a pait of parallel mirrors, one half silver, the other full. The beam of light enters at an angle, and then the position of the output from the half silvered mirror is wavelength dependant. They are used in situations similar to a diffraction grating, but they can be made much more accuratly. (There are other criteria too, which I forget.)
This seperates very close fequencies. Imagine - instead of giving a machine an IP, give it a frequency.
Haveing a laser that is tunable in frequency is not too difficult - the simplest solution would be to use an Optical parametric oscilator. These split the laser into two different colours, and you just block the colour you don't want.
Nevermind then..
Let's just set up a bunch of those dealies in a giant array which has all the functions of the world's fastest punchcard-based computer!
-- 'The' Lord and Master Bitman On High, Master Of All
My figure of 6.6 microseconds is a limit that you can't improve upon no matter how good your technology gets. No experiment yet performed has shown otherwise. A few microseconds is already a long time for modern processors.
The point is that as processors get faster, it will actually become more important to put them close together, if the problem you are working on requires a lot of communication between processors.
The NYTimes article suggests that spreading machines around is the wave of the future. They may be correct so far as flexible "grid" computing (a.k.a., a good old-fashioned multithread-capable batch farm) goes.
But for the hardest computational problems, it will always be highly advantageous to keep the computational power as localized as possible.
Unfortunately light travels pretty slowly.
.. regardless of bus speed ..guess what ... it's a bottleneck!
A 1 Ghz processor does a computation before light can travel 35 centimeters (little more than one foot). So if memory or a second cpu is placed more than a foot away
He may be racist. But judging from this charming subject line, you seem to be a fucking homophobic troll .
Bainbridge Networks:
Brochure of Product
"The Bainbridge Networks XBEAM96 is an all-optical switching fabric that is both protocol (IP, Optical Ethernet, SONET, etc.) and bit rate (OC-3 to OC-768) independent. The defining features of the XXBEAM9966 are its ability to provide per channel, in-service growth and in-service field repair."
Basically they are trying to provide a cost effective way to insert additional switching capability to an existing network.
Beans
This reeks of Smarr, the man who ran an NCSA that did virtually nothing except some flashy demos and Mosaic. Mosaic being a project that was actively discouraged by upper management until it became too successful, whereopon they took credit. Smarr is a master politician, but lacks an eye for people and projects that accomplish something, rather than just looking superficially cool.
Nothing can move faster than the speed of light in vacuum, to be more precise. It's entirely possible to have particles traveling faster than the speed of light in a medium. For charged particles, this causes a radiation field known as Cherenkov radiation, first observed in 1934. It's somewhat similar to the sonic boom observed when something moves faster than the speed of sound.
Andrew S Tanenbaum's book on Computer Networks points out a similar trend; that bandwidth is increasing faster than processor speed. In the future, it'll veyr likely be faster to transfer information about than to process it locally. And that means that distributed computing might become intrisic to most software.
The internet in itself might become a resource for idle CPUs. With a few billion or more individual systems networked up, playing that game of Quake 10 might rely on the processor time borrowed from others.
ya and while their at it, go mode the us mint down too.
I had no idea that Texas is actually in Israel :)
Hmm....
Do we need to sign up for these accounts, or is there a free way in?
Signing up for the NYT account is free!
An easier way is to go to Google News and search for the story. e.g. in this case, search for "texas supercomputer"
You need to take some information theory classes. A light wave is periodic, and a periodic signal is entirely deterministic, and therefore carries no information, other than it's own properties.
They're talking about sensing the beginning of a light wave and producing the peak sooner than it enters. All that is is a phase shift. The only 'information' is the frequency, which you don't need the whole pulse to tell you.
More interesting would be the Mozart example, because it seems to imply modulation. NO details are offered about that experiment, suspisciouly.
One of my buddies used to work for them, before they laid people off in the downturn. They have some really cool stuff in their labs. 32 x 32 Fiber routers that do everything as optically and not electronicly. Probably one of the biggest innovations in fiber optics in a long time unfortunatly it came out when there is a big capacity glut in telecom. At least they got a supercomputer sale, so hopefully they will be able to survive on things like this until telco picks back up.
I beg to differ. I took a class at Oregon Graduate Institute on "Optics for Engineers". One of the two main topics was the use of optical transforms as computation engines. A simple lens performs optical transforms at rates that exceeded the then-total computational capability of the entire planet. The issue is programmability.
It is relatively simple to construct 'computers' for performing transformations from spatial domain to frequency domain and back. There are many algorithms that could be done this way at blinding speed, given the proper inputs etc.
Since that time, various schemes for providing some at least limited programmability have been proposed. For example, a relatively simple 1Kx1K optical filter would be capable of performing 1 million operations in parallel. If the filter used a fast transforming medium like (IIRC) Lanthanum Zirconate - the stuff used in the sunglasses used by the military to save eyes from nuclear blasts), this filter could be changed thousands of times per second.
A similar diffraction-based filter can switch a signal between optical paths, which is the key to providing logic operation. This is similar to what's happening in the optical routers. This can again be done for thousands of signal beams in parallel. So the cycle time is low, but the data rate is huge. In the class we showed that the potential data rate is dramatically higher than any possible digital processor.
The applications for computations of this kind are limited - they're not particularly useful for business data processing - but where they are useful they have the ability to exceed pure digital processors by tens of orders of magnitude. There has even been some thought and work on using hardware like this for certain kinds of database searches.
Sorry I can't give any specifics - I haven't looked at this stuff for a long time.
It's easier to be a result of the past, but more fun to be a cause of the future! http://www.spacefinancegroup.com/
== TrollBuger
(acting like a CHICKENSHIT TOSSER by posting ANNONYMOUSLY!)
Try thinking before you submit! When you said:
you really should have said:
Gezzz what a tosser!
One day the King decided that he would force all his subjects to tell the
truth. A gallows was erected in front of the city gates. A herald announced,
"Whoever would enter the city must first answer the truth to a question
which will be put to him." Nasrudin was first in line. The captain of the
guard asked him, "Where are you going? Tell the truth -- the alternative
is death by hanging."
"I am going," said Nasrudin, "to be hanged on that gallows."
"I don't believe you."
"Very well, if I have told a lie, then hang me!"
"But that would make it the truth!"
"Exactly," said Nasrudin, "your truth."
- this post brought to you by the Automated Last Post Generator...