Internet2 Speed Record Broken

RevKa writes "InternetNews.com has a report of a new Internet2 land-speed record. The old record was nearly cut in half: the two parties, California Institute of Technology (Caltech) and the European Organization for Nuclear Research (CERN), 'transferred 859 gigabytes of data in less than 17 minutes.' InternetNews goes on to say, 'This record speed of 6.63Gbps is equivalent to transferring a full-length DVD movie in four seconds.' Various scientific purposes were mentioned 'as well as commercial applications from entertainment to oil and gas exploration.' The article ended with hardware specs 'S2io's Xframe 10 GbE server adapter, Cisco 7600 Series Routers, Newisys 4300 servers using AMD Opteron processors, Itanium servers and the 64-bit version of Windows Server 2003.'"

Comments

wow

[Anonymous Coward]

how much bandwidth does doom3 need for network gaming?

Re:wow

[Lt Cmdr Tuvok]

how much bandwidth does doom3 need for network gaming?

It is typical of humans to focus primarily on the ways in which new technology can be utilized for 'fun'. Computer games are a particularily ubiquitous example of this phenomenon. Massively networked computers have the potential to become the greatest compound computational device that mankind has ever had access to. If only the proper effort were expended, multiple paralell processing tasks could quite easily be run on this supernetwork. The combined power of this cluster would thus be beneficial to all.

There is slim hope that this will happen, at least in the foreseeable future, human logic being as flawed as it indeed is.

Re:wow

[AviLazar]

Dude...your Vulcan speak is freaking me out man... ;)

Re:wow

[qray]

Definitely beneficial! We could take Tribes 2 or Quake, or the like and build entire simulated armies. Instead of actually killing people we could simulate wars and just abide by the results.

And you thought computer games were bad. They may save us from extinction.

Re:wow

[Jameth]

"There is slim hope that this will happen, at least in the foreseeable future, human logic being as flawed as it indeed is."

Ah. So, what will this superior form of logic gain us? With a super-efficient system we could solve all sorts of problems and extend our lives and enrich ourselves, allowing us to have longer to enjoy...wait a minute, you're complaining because we'd rather be able to enjoy ourselves, which appears to be the point anyway, than to not enjoy ourselves for a while so that we can later enjoy ourselves as we would have been doing anyway.

Perhaps you could explain your 'unflawed logic' sometime?

Re:wow

[Illserve]

Oh Cmon.

Giving the average person access to a "compound computational device" would be about the biggest waste of resources in human history.

Re:wow

[antic]

"We could take Tribes 2 or Quake, or the like and build entire simulated armies. Instead of actually killing people we could simulate wars and just abide by the results."

Great idea, but only if Diebold gets to orchestrate the simulation...

Re:wow

[cpghost]

Who said research wasn't supposed to be fun?

Re:wow

[Not_Wiggins]

Welcome to "Skynet." 8)

Re:wow

[drinkypoo]

Parallel computing is the forseeable future of computing. If you have sufficient bandwidth across the internet, it only makes sense to share computing resources between people. Just think about how it could help gentoo users :)

Station wagon full of backup tapes

[RollingThunder]

I think we've well surpassed what a station wagon full of backup tapes can do now....

Re:Station wagon full of backup tapes

[IWannaBeAnAC]

I don't think so. One DVD in 4 seconds... how many DVD's can you fit in a stationwagon? A few thousand at least (probably 10's of thousands). Each thousand DVD's gives you an hour to drive to the destination to meet the bandwidth requirements.

Except, that I don't think you can drive from CERN to CalTech, even with a few days to do it ;) So, you might actually be right! But they still have a tremendous way to go to exceed the bandwidth of a supertanker....

Re:Station wagon full of backup tapes

[real_smiff]

supertankers are very slow. they have very bad latency. and do these DVDs have cases? :p

Re:Station wagon full of backup tapes

[Rostin]

I think that to be fair, the DVDs would have to be burned at the point of origin and then read at the destination.

Re:Station wagon full of backup tapes

[nmk]

That may be true. However, how long do you think it would take to burn those tens of thousands of DVDs that you're going to transport in the station wagon.

Re:Station wagon full of backup tapes

[IWannaBeAnAC]

Yes, but that can be massively parallelized. The total time taken need only be the time to burn & read 1 DVD.

Re:Station wagon full of backup tapes

[Gunzour]

From a business perspective, think of how this could change distribution. Instead of having to ship DVDs to your local Blockbuster, Blockbuster could have their stores hooked up to the net and just download the DVD's their customers request.

"Oh, no more copies of Passion of the Christ on the shelf? Hang on a couple of minutes while I burn one for you..."

Re:Station wagon full of backup tapes

[zijus]

Hum... Not really "back-up" though. The kind of data which are gonna be produced at CERN in LHC won't be backed up : just too big.

The point of those super fast transfer rates is that probably experiment data will _not_ be stored at CERN at all, just produced and sent straight to storage farmes. Hopefully organisation will be clean enough to avoid more transfere after first storage. The goal beeing : stored where it's gonna be analysed.

I worked a while within the ALICE experiment off-line software section at CERN in 1999-2000. The proportion of problemes are just mind blowing. Some experiment generates so much data at each run that a single extra bit used to store a raw data, immediatly translates into 10th of extra disks required. One run is at max a few seconds. As fare as I remember, ALICE ( A Large Ion Colider Experiment) will generate on it's own 1PB a year. The why of it is: as opposed to colide e-/e+, this experiment will colide nucleus. We could see it as: colide two car side mirors and you get X by-products. But now, colide not the miror only but the whole cars: you get magnitude of order more by-products to observe. The goal , as fare as I rember is to observe some Quark-Gluon-Plasma (QGP), a gas of quark and gluon. It's the fifth state of mater we know about.

Some lads in slashdot thread mentionne p0rn... But just a single peta byte is equivalent of what has been said and written ever by humanity. We'd have to generate a lot of pOrn/mp3s/... to have that much data. Or maybe set up ALIPE ( A Large Intricate P0rn Experiment.)...

Z.

Re:Station wagon full of backup tapes

[Teun]

Won't work, CERN is European DVD region 2, Caltech is North American DVD region 1.
You can tipically only change this code 5 times before it locks up.

At least that's what the *IAA's would like us to believe :)

Re:Station wagon full of backup tapes

[ivow]

What about filling up the wagon with fiber channel SAN's?

btw: I think we're forgetting that fold out seat in the back of the station wagon.

This sounds great!

[jkrise]

Most desktops don't have that much bandwith on their FSB!!

Re:This sounds great!

[GreyPoopon]

Pray, what's the point in adopting a standard today, that most common devices that need internet access (read PCs) can't even dream of attaining?

Remember that this is an experiment, and getting speeds like these into widespread availability is pretty far in the future. By the time such speeds are available, the computing power to take advantage of them probably will be too. If they don't start the research now, we'll have very powerful computers that come to a screeching halt everytime they have to retrieve data from the 'net.

Re:This sounds great!

[Anonymous Coward]

Well... Let's see, an LTO-2 drive can spool at 35GBps (that's bytes not bits) a StorageTek L700 can house 20 of them, that's 700GBps (which is approx 5600Gbps) your LTO-2 tape can store 200Gb native, err... I'm getting boared with the maths now, but you get the idea...

Re:This sounds great!

[afidel]

ERROR: Order of magnitude problem
With a transfer rate of 60 MB/sec, the Ultrium 460 is the ideal choice for enterprise-class data protection needs. linky

So, real numbers are max 1.2GB/s or 12Gb/s for the L700, not bad, but not that much faster than this transfer. And with the tapes you still have to transport them to the destination to make the comparison fair.

DVD speed

[Gunzour]

'This record speed of 6.63Gbps is equivalent to transferring a full-length DVD movie in four seconds.'

Yeah, that's the message we want to convey to the MPAA. Everyone knows the Internet2 is all about pirating DVDs.

Re:DVD speed

[hasdikarlsam]

Who cares what it's good for?

I want one!

Re:DVD speed

[jkrise]

Think how much faster we can get our Service packs from Microsoft! Download in less than a second - but rebooting would take ages..... groan ;-(

-

Re:DVD speed

[Mad Leper]

Even better, the fact that this is not the first article to refer to network speeds as "DVDs per second/minute/hour" It must piss off the MPAA and RIAA when the media uses DVD/MP3 as a benchmark for network performance.

Re:DVD speed

[hype7]

Yeah, that's the message we want to convey to the MPAA. Everyone knows the Internet2 is all about pirating DVDs.

Someone give Jack Valenti a call! His exit interview was linked off here just a few days ago, and he said:
If everything stayed just as it is right now, we could probably survive it, because even with broadband it takes at least an hour to bring down a movie. But I visited the labs at Caltech, and they're running an experiment called FAST where they can bring down a DVD-quality movie in 5 seconds. The director told me it could be operative in the market in 18 months. Well, my face blanched.

I wanna know what his face does when he finds out we can now do it in under 5 seconds :D It sure couldn't get any uglier than it already is.

Anyway, don't let him quit before someone tells him!

-- james

Re:DVD speed

[darc]

>'This record speed of 6.63Gbps is equivalent to transferring a full-length DVD movie in four seconds.'

Sheesh. Whatever happened to the last benchmark unit? Libraries of Congresses? All you kids and your new fangled metric system... DVD units. Back in my day, we were sued by BOOK publishers! Not some crazy eight track industry. Those were some REAL copyrights.

*prattles*

not bad...

[Slashbot Hive-Mind]

here are some other records (taken from here:

Current Records
IPv6 Category

Single Stream Class: 46,156 terabit-meters per second by a team consisting of members from the California Institute of Technology (Caltech) and CERN across 10,949 kilometers of network.

Multiple Stream Class: 46,156 terabit-meters per second by a team consisting of members from the California Institute of Technology (Caltech) and CERN across 10,949 kilometers of network.

IPv4 Category

Single Stream Class: 69,073 terabit-meters per second by a team consisting of members from the SUNET, the organization for the national higher research and education network (NREN) of Sweden, and Sprint across 16,343 kilometers of network.

Multiple Stream Class: 104,528 terabit-meters per second by a team consisting of members from the California Institute of Technology (Caltech) and CERN by sending 859 gigabytes of data across 15,766 kilometers of network in 1037 seconds (just over 17 minutes), for an average rate of 6.63 gigabits per second.

Re:not bad...

[drinkypoo]

That's interesting, for IPv6 the single and multiple stream records are the same, but for IPv4 the single-stream is considerably slower. I'd very much like to see an explanation of why this is so.

In tomorrow's news ...

[sawb]

.... the RIAA and MPAA sue Internet2 as being a potential source for copyright violations by being able to steal a movie in 4 seconds or an album in 0.0003 seconds.

Re:In tomorrow's news ...

[jkrise]

and users of internet2 would be sued for possessing the equivalent of 6.83 billion CD writers!

Re:In tomorrow's news ...

[tepples]

"A tax of $0.01/MB will be levied on all network transmissions except those originating directly from our licensed content distributors."

Of course, if the country has any sort of constitution guaranteeing freedom of speech, copyright law will allow any copyright owner to join the royalty pool. This is already the case with sound recordings and blank Music CD-R media.

Oh-oh.

[King_of_Prussia]

'This record speed of 6.63Gbps is equivalent to transferring a full-length DVD movie in four seconds.'

Explaining it like that is likely to draw the wrong sort of attention - How long until Jack Valenti and his crew of RIAA/MPAA thugs descend on this new menace to their livelihoods?

Incidentally, for some reason gmail has decided to give me 12 invites - they will go to the first 12 logged in posters telling a funny joke involving ESR or RMS, bonus points for use of ASCII.

Re:Oh-oh.

[meringuoid]

Incidentally, for some reason gmail has decided to give me 12 invites - they will go to the first 12 logged in posters telling a funny joke involving ESR or RMS, bonus points for use of ASCII.

I think gmail must have just hit some kind of critical density; I finally got my invite last week, and since then I've been seeing them offered just about everywhere.

That's exponential growth for you, though... I've invited three people in myself already, and I imagine they've got invites of their own by now. I doubt Google will ever need to officially open gmail to the general public; in a couple of months the number of invites in circulation will probably exceed six billion :-)

Evil plan

[Libor Vanek]

1, Manage to transfer your data at 6.63Gbps
2, ???
3, Profit!

Windows..

[Quixote]

Why don't they do this test with an OS like *BSD (or Linux), with its highly-tuned networking stack?

Re:Windows..

[dk.r*nger]

Why don't they do this test with an OS like *BSD (or Linux), with its highly-tuned networking stack?

Because Microsoft has a marketing budget and Caltech/CERN don't give a rats ass what software it runs when it's the network infrastructure they're showing off..

Re:Windows..

[fdisk3hs]

The last record a few months ago was set on NetBSD. It's a game of leapfrog.

Distance is as impressive as speed

[erick99]

The distance of approximately 9,800 miles is as impressive as the speed. The article did not mention how many devices (i.e. switches, gateways, etc.) that the data passed through from site to site.

Cheers,

Erick

joke, best I could do on the spur of the moment..

[sgtron]

RMS walks into a bar. Bartender says "Hey, we don't allow hackers in here."

RMS Says "Huh.. that's GNU'S to me."

Interesting point...

[Noryungi]

This is straight from the article:

Internet2 is fast -- Abilene, a U.S. cross-country backbone network, blasts data at 10Gbps. But transoceanic networking is another story. There are hardware and software issues to overcome, Gray said.

For example, one limiting factor is that the fastest available interface for PCs is the PCIX64 Bus Isolation Extender, which can only handle 7.5Gbps.

So... Let me get this straight... The problem these guys have is that they are using PC to connect to, and send data on, Internet2?

I remember a time when "serious" CS researchers would not touch a PC with a ten-feet pole. Times have changed, indeed.

Re:Interesting point...

[mbbac]

Would a Next Cube be considered a PC? I would, because it's present day brother would be my PowerMac G5. And a Next Cube is the PC that Tim Berners-Lee used when developing the Web.

Re:Interesting point...

[RupW]

I remember a time when "serious" CS researchers would not touch a PC with a ten-feet pole. Times have changed, indeed.

Because advancement is market driven and PCs are where the money is. That's probably the fastest price / performance bus they can get. Research institutions aren't made of money (unfortunately).

100 Mbps

[bob_avernus]

This makes even the Japanese and Korean connections http://slashdot.org/article.pl?sid=04/09/02/132221 5&tid=95/ look pitifully slow. Just what we need the ability to watch 30 high def soaps at once...

Err - cancel that

[defsdoor]

Just realised the file was in our proxy cache!!

Achieving equivalent Disk I/O

[digid]

What kind of equipment is needed to achieve the necessary Disk I/O to match the network throughput?

Re:Achieving equivalent Disk I/O

[burns210]

RAM... lots and lots and LOTS of RAM.

Re:Windoze??

[melkorainur]

The article doesn't seem to say what protocol was used. I was hoping to find out if it was TCP or TCP based like iSCSI or something lower level. If it's lower level. The second thing is about the s2io card. I'd like to know if it has true TCP offload capability like stateful packet handling and processing, ie: strip headers off at NIC level and dma to host memory. Even S2IO's site doesn't elaborate on that. Anyone know?

Give me mass storage of that throughput...

[Vo0k]

I'm sick of waiting 2 mins to transfer a DIVX movie to a different partition.
For us, average nerds, if we ever got connection that fast, it would still feel slow because of our storage speed. :P

Re:Windoze??

[MagicBox]

Maybe this will help

PCWorld

Looks like they were using next-generation Internet Protocol version 6 protocols, but I am not sure if that encapsulates some next generation of TCP/IP as well

Wheelchair

[Himring]

Itanium servers and the 64-bit version of Windows Server 2003

This reminds me of another article this week where a guy strapped jet engines to a wheel chair....

Benefits?

[shadowkoder]

I would like to know the benefits this sort of bandwith testing brings about. Does it help determine bottlenecks in current technologies? Help determine roadmaps for future techs? Or is this just some testosterone releasing between researchers? :)

Use *NIX and it would be

[HermanAB]

about 8 minutes.

I guess they wanted to leave themselves some room for improvement and therefore started off with Win2003...

Point of life is fun

[dunc78]

Seems to me that everything we do in life it to have more fun. Why is technology beneficial to us, because it makes our lives more "fun" or gives us more time to have fun. If it weren't fun why would we want all this technology to extend our boring lives.

Re:What about LOC/s?

[maxwell demon]

Sorry, but Google apparently doesn't know the unit Libraries of Congress per second.

I can beat that!

[Mateito]

Look, here I have a 80GB HD full of... um.. arthouse movies. I pick it up, move it a meter. Takes around half a second. Thus I am moving effectively moving data at 1280Gb/sec.

Beats their record.

Oh? It needs to go over wire. Fine.

Be amazed at my 80GB Harddrive over Cat 5 FLYING FOX!!

Bwahahaha.

It's for backbones.

[christophla]

This type of speed is for shared backbones - improving qos for tens of thousands of users at a time. You're not going to get these speeds between two endpoints.

anyone remember those 3dfx commercials?

[Trepidity]

3dfx had some commercials a few years ago that had a documentary-style voice talking about how advances in technology could help create a better tomorrow, more efficient farming that would feed the world, and so on, and then some guy says "hey, or we could use it for games!", and then they introduce the voodoo-whatever-number-it-was-at-the-time.

It's fast, but...

[Spad]

Sure it's fast, but it's not that great. SuperJANET 4 is running on a 10Gbps backbone with plans to increase it to 20Gbps in the near future.

There's nothing quite like having a 2.5Gbps net connection coming straight into your department at uni :)

Internet2 Speed Cop Pulls Them Over

[red floyd]

Ok boys. Did y'know that y'all were doin 6.63Gbps in a 5.5Gbps zone?

I'm afraid I'm goin' t'have t'write y'all up for speedin'.

Scalability?

[Thedalek]

These Internet speed record experiments are interesting, but the issue of scale is rarely addressed. Okay, so a team of researchers were able to go faster than the speed of bad news, but what happens when the server load is a bit higher than just one transfer?

Or does Internet2 use some exotic de-centralized transfer method that renders the paradigm of servers laughably obsolete?

What about Latency?

[closms]

I agree that bandwidth is great, but applications like distributed file systems are much more sensitive to high latency. Any stats on Internet2 latency?

they needed data

[twitter]

Why don't they do this test with an OS like *BSD (or Linux), with its highly-tuned networking stack?

They needed data. They started with DVDs they owned, but a few dozen only added up to about 1/8 of what they wanted. Renting was too expensive and they were worn out from ripping the first 12. The solution was obvious ...

The connected the Winblows 2003 server and used it to collect data. Within minutes, it was rooted and it's reputation for good network connectivity spread quickly. In a day or two, the multiple terabyte array was filled with music, movies, porn and warez. The data was then transfered to reasonable hardware and the test was performed.