Slashdot Mirror
Terabit Routers
Rocket Boy writes "I was perusing the news and came across this sucker. The specs on the thing look outrageous. Heck, the whole thing looks outrageous. 2.5-5.6TB/S speed, Supports 2240 OC-48 or 560 OC-192 connections. "
You can download a lot of po.. I mean play a lotta qua... I mean
read a lot of slas... I mean.. work. You could do a lot of
work with that.
you can't afford it.
In most cases, I beleive the reason we /. a page is not the bandwidth, but the processing power. Handing out all that info to a bunch of 28.8's...
> it would be nifty if ALL backbones used these
Sheeyah. Good luck finding a telco that can offer you enough hose to fill even 25% of one of those gadgets.
This product isn't about plugging OC-192s into the back of a single host, it's about aggregation.
A.C.
DDE was President when the Interstate Highway legislation passed. It was sold as a defense project. Boulder Dam was started under
Hoover -hence the name. (FDR tried to change it to Boulder dam)
"You can put DOS on the network"?! My network administrator
We're not forthcoming because we have a patent in process on our architecture. Though it does resemble a hypercube! ;-)
Veni Vidi Avici!
Then see the black helicopters flying over your house, and the tactic squad smashing your front door yelling at you to get down on the floor. Really awesome, isn't it?
It would take a finite amount of time for the packets to reach the machine being /.'ed.
$80K per port.
2240 x 80K...you get the idea.
ummm... you must be stupid.. remember that the major (well when it was designed) use of the modern backbone was for mear 4KHz fone conversations... and if you take the amount of bandwidth in an oc-198 and divide that by 4KHz (or the equiv value in KB/sec).. i think you will see that was QUITE a bit of calls it could handle.. and as the INet grows (such as it is), the lines WILL need to be expanded.. but for now it suffices and change isn't something that occures at once.
All the engineers left Cabletron three years ago.
I hope you're not in a hurry.
Somebody should put together a beowulf cluster of these.
What? Why is everybody rolling their eyes? What did I say?
Why waste all that processing power rendering a billion FPS? The human eye can't process images that fast.
Use it to render about 80 FPS with full radiosity and objects with complexities of >100k triangles!
More like 11 or 12 seconds to fill your hard drive.. assuming its some kind of fiber channel array you're writing the data to. ;-)
Posted by FascDot Killed My Previous Use:
/.'ing resources."
..."assuming we have an infinite amount of
well would be nice and fast assuming EVERYONE had that kinda connection :P what we REALLY need is subspace relays ala star trek ;) remember that the computers on star trek have a subspace field around them...so you think 1GHz is fast? their computers run at Warp 3.36 (~50c if i recall)
:P
:)
okay enuff of my foray into star trek land
these terabit connections are tres cool
"There is no spoon" - Neo, The Matrix
All I can say is "wow." I want an oc-192 connection! (fill up my 9GB harddrive in .1 seconds or so) ;-). Well, my dream will (probably) never be realized but if any of you out there want to donate one to me call...
My plan is to pimp before they realize I'm a jackass. Hit 'em hard and fast.
or rather, it would be nifty if ALL backbones used these. But with all these OC connections you run into a problem being discussed over on the IPv4 thread, not enough IPs to go around. But all problems solved it would be a rad system. But if you had a terabit network hooked up to anything smaller than a super computer it would explode. A gigabit network comes real close to maxing out your average PCs bus architecture terabits would make my little 66mhz bus melt :(
I'm a loner Dottie, a Rebel.
If you are insane..
Quake III (if it was built for SMP) @ 1 billion+ frames a sec.. hahaa..
"Windows 98 Second Edition works and players better than ever." -Microsoft's Home page on Win98SE.
I ate my tag line.
-=Ellis (D)25=-
Would need to launch a smurf.. Just lauch out a ping that grows in size over time.. Take out the first site, then the next.. Just take out a nice back bone or a 100...
"Windows 98 Second Edition works and players better than ever." -Microsoft's Home page on Win98SE.
I ate my tag line.
-=Ellis (D)25=-
.=blanks
W=West
C=Central
E=East
H=Horizontal
W....C....E
XHHHHXHHHHX
W....C....E
"Windows 98 Second Edition works and players better than ever." -Microsoft's Home page on Win98SE.
I ate my tag line.
-=Ellis (D)25=-
They need a Mae Central and a Mae Horizonital also if you want a good backbone system... It would look like this in ascii
I I I
I----I----I
I I I
"Windows 98 Second Edition works and players better than ever." -Microsoft's Home page on Win98SE.
I ate my tag line.
-=Ellis (D)25=-
6 terra.. Hmmmm.. It would be hard to /. it. You would have to exploit it.. /. could generate enought trafic to knock off 1 terra...
I don't think all the user that read
"Windows 98 Second Edition works and players better than ever." -Microsoft's Home page on Win98SE.
I ate my tag line.
-=Ellis (D)25=-
There's actually 4 terabit startups:
Avici - Has several betas ongoing. I know the product works, cause I work at Avici.
Nexabit - Close to releasing product, but it uses a crossbar arch., not as scaleable as ours.
Pluris - Claims similar to ours, though they haven't shown anything.
Juniper - Has a gigabit router now, working with Crisco to get a terbit unit.
You should check them all out. But we do have units working, and we have played a few mean games of Half Life on these beasts. But to be sure, these are the units that will be placed in phone and cable POPs to increase bandwidth on the Internet backbone. At a cool 1 million clams a piece, I wouldn't go yanking up CAT5 just yet.
Who cares how fast routers if:
(a) not everyone is using them [the 'net is only as fast as it's slowest link]
(b) it's all gotta squeeze through your modem at the end of the line.
We need a massively ambitious program (like the interstate highway program) to get high speed data access to every home in the nation. Damn the costs, the new 'net will generate new revenue to cover the costs down the road.
The site said that the routers were scalable to terabit speeds, but that doesn't mean that a terabit router has ever been built. It would be possible to put to take a couple million CPUS and build a petaflop supercomputer, but no one could possible afford that (at least not yet).
It is interesting to note that the interstate highway system was also built as a backup runway system for the US military.
Evidence: interstate highways, especially the early ones, are mainly straight with few turns, have trees far back from the side of the road to accomodate wide airplane wings, are divided to provide a parallel road for military traffic, have controled access and are thus easy to close off.
2^5
Posted by Mister_BFR:
The Nexabit team, somehow, has managed to make a
name for themselves as hypemasters. They have
squeeged out press releases that make outrageous claims. They even went so far as to make a press
announcement using Frontier Networks name without first getting their permission. There are obvious legal problems with doing so, but even more so, they don't even have a working module in the lab at Frontier. I've talked to them all, and the only company I know of - to date - that has a working architecture, in fact, a kick ass architecture is
Avici. They based their architecture on supercomputers and they stuff up to 560 modules in a single monolithic architecture that scales. They did it right. First they built hardware that performs and scales, then they built software on top of it that rocks... if you saw their operating system CLI, you'd really like it. SO, in this crazy field, I can promise you that these guys are
REAL and it WORKS.
Posted by Mister_BFR:
Avici has done it. Unlike their competitors, the way you calculate total bandwith with Avici is simple. Mutliply the # of interfaces, by the bandwidth speed and you got it. It's that simple. Every port is wire speed whether you are running at OC3 or OC48. Unlike NExabit (the breakfast cereal), who calculates the bandwidth off of some derivative of backplane specification Avici over-engineered their box to be NON-BLOCKING through the use of virtual channels and dedicated buffering. And, by the way, Avici is NOT the only company that said that they to scale the system to terabits that you need to add more frames... in fact, SHOW ME ONE VENDOR WHO can scale to 5.6 terabits in a single frame - jeez.. no one can, it's physically impossible today based on today's electronics... and even if they could, it would have to be a water cooled unit to handle the heat. As for your comment on off-bus memory access - it shows that you have not had a single technical presentation on the product. If you are truly interested, i suggest you contact Avici and arrange it BEFORE you pontificate about an architecture you clearly know very little about.
And besides, the Avici story works.... I've seen it. Avici is a name you will be hearing about for a long time to come.
Maybe we should wait until the current boom is over and we start sinking into recession. Something like this could create a lot of jobs and involve a lot of government spending, which is what is needed to end a recession. Then we may just start another long boom.
It's not enough to bash in heads, you've got to bash in minds. - Captain Hammer
Say, suppose some webserver were connected to obscenely gross bandwidth: how long would it take us to /. it? (assume that we have an infinite amount of /.'ing resources).
In Soviet Russia, Jesus asks: "What Would You Do?"
In Korea, some highways are closed once a year for landing practice. They're similar to US interstates, but they have movable median dividers.
I don't think the US system is much use as a landing zone. Landing is easy - it's finding gas, ammunition, and rental cars that's difficult.
---- "If we have to go on with these damned quantum jumps, then I'm sorry that I ever got involved" - Erwin Schrodinger
A bit faster than "a few years", and Internet backbones get them first - cellphone networks actually have fairly low traffic flows.
You'd want a cluster of routers of that capacity at each physical POP that's connected by high-capacity fiber. Those will connect to metropolitan-area rings, and those to businesses and ISPs.
See the BBN^H^H^HGTE Internetworking fiber map.
-jwb
Kremvax was named by Piet Beertema long before any russians joined usenet/internet.
x .html
http://www.lysator.liu.se/hackdict/split/kremva
HTH
Tob
The hardware isn't the only problem. We're using 2 Cisco 4700 in our university backbone and our traffic to the outer world increased a lot (2.5 times more than 2 weeks ago because of a better connection) and so did our internal traffic. What happened ? Our network crashed every other day, no clue where the problem was (we suspected hardware failures). No we found the solution: a software bug in the cisco's (they're running at about 60% the speed they claim they could handle...). ;-) )
So, tech-specs are good, but you have to test them in a real world enviroment (ok, we're only a university
OC-48 and OC-192 sounds nice, but I'd be happy, if
OC-3 Modules for the Cabletron SSR-8000 are available 8-(
Just for the record, I used 9600 :o)
---
How 'bout one of those dummies book..
Beuwolf Building for Dummies.
"Windows 98 Second Edition works and players better than ever." -Microsoft's Home page on Win98SE.
I ate my tag line.
-=Ellis (D)25=-
There are many more differences between OC-3 and OC-192. It is true that 192 can be broken down into OC-3 (and OC-12/OC-48) streams, but it's not done by sticking OC-3 streams out on different wavelengths.
:-)
The internal design of the SDH/STM networks is pretty interesting, but as always the problem is at the switches. The major problem with gigabit+ routers is always getting stuff from one port to another. The subsiduary problem is what to do if all the incoming traffic needs to go out one port.
Ah, the joys of router design. I do miss it sometimes
Life would be a lot better in the USA if the guys at MaeWest decided to implement one of these suckers there. That one point causes a massive amount of net traffic in this nation.
RB
Taking this to the logical extreme, we ought to have one of these in each home around..
break out the envelope:
assume 56k now, doubling every, what 1.5 years? that would 1tb in about 15 iterations, 15*1.5 = 22.5 years. I can wait for that.
I see it a lot. Everybody is saying that a PC is the solution for everything. Need a high-end web server? Use a bunch of PCs! Need a router? A PC will do your job!
I disagree with this. Perhaps another type of system, but PCs are probably the worst and most inefficient architecture, mostly because they remain to be the only existing CISC architecture today, whereas everybody else is using a RISC processor.
I also would bet that all the PCs you need to equate to the product mentioned above would cost way more, and would take much more time to maintain.
On the issue of scaling in particular, I think this product was made with that bandwidth in mind, if not, what's the real use in marketing it as such?
Great, bring up the chicken or the egg whydontcha.
/. effect. I highly doubt that most /.'ers are on a dedicated T1 to the net but on a 28.8k+ to WaveModem & we still manage to bring down servers that are on dedicated T1/3's, etc...
Don't forget you've got 100's/1000's/etc... of people on the end of a line using that bandwidth, not just one squeezing at it.
Take the
If they terebyte xfer ability was there we'd upgrade dontcha think? Mbit ability has been available on the net for years & we're finally catching up to it. (2400, then 14400 (no one used 9600) then 28.8, 33.6, ISDN/ADSL or 56k OR WaveModem, and we're still not as fast as what the hubs & routers are talking to each other at AND we still think its slow.
We have the reverse situation at my company. We have 10/100's in everyone's PC but the routers & hubs are only 10's. What a pain! There's only 250 PC's systems to connect...they should get with it.
Anyway, up here in Canada Shaw has been VERY ambitious about wiring us up with the Wave & the ISP's are advertising *DSL lines. I don't know of an area in Calgary that doesn't have the wave. Maybe you need to move?
--Clay
So anyway.... The only difference between OC-3 and OC-192 is the number of seperate light frequencies that are sent across the line. Fact is that most fiberoptics can handle a considerably larger number of light frequencies than is currently being done. The only problem is developing a switch/hub/router which can decypher all of these seperate frequencies. It's not all that hard, but it's damned expensive. There's no point in teaming together 2000 OC-192's when you can just create better technology which will be just as fast... and probably end up costing less, and be a bit more reliable. I can't image how much CPU time it would take to route 2000 different connections... that's mad.
Looks like the Dense Wave Division Multiplexing technology is going to take over in a few years. I bet you'll see this technology appearing in digital cellphone networks, followed shortly by inclusion into the Internet backbones.
A couple of these routers stationed across the world should handle an enormous amount of bandwidth.
If you take a look at the whitepaper, you'll see that they're partnered with Nortel-- one of the leading telecommunications companies.
What an amazing product.
æeee!
If you are looking for some other bad boys there are plenty of other startups that are doing the same thing. As I recall, Nexabit plans on having a set of boxes that you can link together and route up to 64 OC-192c streams. It might be more. Check these places out:
http://www.nexabit.com/
http://www.argon.com/
http://www.ironbridgenetworks.com/
I'm sure there are others I'm missing.
nick
This box is not revolutionary. Current switch fabrics already are set to do OC192, and vendors are shipping boxes with 10-12 slots. Avici is just building a full-height-rack monster with 40 slots. Notice that to get the full speed they state, you have to install a bestiary of 14 full racks. All slots, and the switch fabric per-port speed will still be OC192. Their claim to be able to extend the switch fabric to inter-node connections is dubious. There is no way off-bus memory accesses will happen without significant latency.
As for their hype about Packet-Over-Wavelength, any box that can do OC192 *does* this, as OC192 is the input to a DWDM Transponder.
Pluris was doing it (or claimed to) since '97. They used hypercube for internal switching fabric configuration, and I believe used wireless very-short-distance (1m) links to reduce amount of wiring. Vadim Antonov was the brains behind it, he left the company in '98. Pluris didn't release anything, and their website (www.pluris.com) seems down. Vadim also is person who named major russian UUCP host as 'kremvax', developed/ported unix for soviet PDP-11 clone (called DEMOS), founded first russian ISP (also called DEMOS), architected much of SprintLink, etc.
:)
Avici on other hand doesn't really say much about how their fabric is done, but I imagine it won't be crossconnect (2000x2000 wires...ugh). Probably some sort of hypercube or selective mesh. I wish Avici was as forthcoming with technical details as Pluris.
Saying something is "scalable to X terabits" doesn't mean much unless you specify what an "X-terabit router" really has to do. A big $#@! pile of PCs is "scalable to X terabits" if they're all handling totally independent traffic flows and never have to talk to one another, for example. Probably the most useful measure is "cross-sectional bandwidth" (i.e. the amount of bandwidth you'd have to remove to partition the nodes into two or more isolated subsets) as used for measuring intramachine interconnects, but - alas - Avici doesn't give us that information.
There's also this little issue of balance. Nothing scales perfectly, and often you don't know where the bottleneck will be until you build one. Sure, if you add up all the links maybe you get up to X terabits, but maybe node-internal contention for some resource limits you to X/100. Of course, this never stopped marketing types from acting as though their machine/link/router would be the first in the history of computing to scale perfectly.
Slashdot - News for Herds. Stuff that Splatters.