x
Routing

Chiaro Girds 'Router' for the Grid

Can the grid-computing craze accommodate a new optical switching and routing startup? Chiaro Networks thinks so.

Chiaro this week said it has shipped a working optical "router" to be used in the OptIPuter, a five-year, $13.5 million grid computing project funded by the National Science Foundation (See Watch for the Grid). The OptIPuter, a joint venture of the University of California at San Diego and the University of Illinois at Chicago, is a network of several computing systems that are linked by optical fiber.

Grid computing systems such as OptIPuter allow their resources to be shared by all the network's end users. If you think of a grid as one giant computer stretched over a large area, Chiaro's system and the optical fiber connecting to it would be analogous to the bus that links the storage, processing, memory, and other computing elements together in a PC.

Chiaro's initial prototype isn't really a "router" per se, but rather a giant, super-fast optical packet switch. Initially the company hopes to simplify service-provider networks by introducing a core optical system that's so fast and so scalable that it eliminates much of the need for an aggregation layer between the network's core and its edge.

Chiaro says it has a switching technology, called Optical Phased Array, that helps it switch a large number of ports at nanosecond speeds. "What seemed to be mutually exclusive in the past was both large port counts and ultra-fast switching speeds," says Ken Lewis, Chiaro's CEO. Lewis claims the company's routing platform, called Enstara, is a step beyond today's optical switches, which handle a high number of ports at millisecond speeds, or today's IP routers, which switch a smaller number of ports at nanosecond speeds.

Here's an oversimplification of how the company's optical switch technology works: It takes light from an input fiber and sends it through 128 gallium arsenide waveguides (see Arrayed Waveguide Gratings (AWGs)). As the light travels through the waveguides, a voltage is applied to control or “bend” the light. As the light comes out the other end of the waveguides, the light enters a five-inch air gap, where constructive and destructive interfering patterns occur.

This constructive and destructive interference is akin to dropping two stones in a pool of water and watching each set of ripples. Some of the ripples will combine and make bigger (constructive) ripples while others will (destructively) collide and cancel each other out. [Ed. note: we had to dust off our high school physics books for that one.]

Anyway, the voltage controls the location of the constructive interference, which in turn creates bright spots that allow the light to be steered and aimed at any one of the 64 outputs on the switch module and -- voila! -- the light is switched in nanoseconds by a machine with no moving parts.

Chiaro's 64 x 64 optical switch module was tested by AT&T Research Labs, as noted in a paper published this month at the Lasers & Electro Optics Society (LEOS) annual meeting in Scotland. "We measured the [Chiaro switch module's] transmission performance with the data rate and channel count per fiber scaled to 160 Gbit/s (40 Gbit/s x 4 wavelengths)," wrote the paper's eight authors. "For a fully loaded switch, this capacity would translate into a potential non-blocking switch element throughput of (64 x 160 Gbit/s) or 10.24 Tbit/s."

"Switch operation is based on fast electro-optic effect, and measured switching time was better than 20 nanosecond, sufficient for IP format packet switching," the paper states.

The Chiaro system being installed in the OptIPuter program is running each port at 10 Gbit/s, and it's not equipped with hundreds of ports, Lewis says. But because it switches optically, he claims, the rate and format of the payload don't matter to the optical switching system. In theory, Chiaro's switch can operate at a much higher data rate and can handle four colors of light coming into each port, which would scale the system's overall capacity significantly.

Chiaro is also talking up the technology that protects the system, saying it will allow for hot capacity upgrades and expansions. In lab tests, Chiaro says its system has been able to maintain "state" with hundreds of peers while failing over from one master control processor to another without losing packets or causing a network outage.

Interesting? You bet. Commercially viable? We'll see.

There's a question of how many service provider networks worldwide would actually need, buy, and use a switching platform of the scale Chiaro is hinting at. Call it what you will -- optical router or optical switch -- but neither market is doing well right now, and the core router market is fiercely competitive (see Router Vendors Look for Bottom).

Chiaro has beaten the odds so far, just by being able to get a product in use in such a visible project as the OptIPuter. Optical switch makers such as BrightLink and optical routing vendors such as IP Optical Inc. both had high hopes and interesting technology, but didn't survive quite so long (see Brightlink Works on Its Grooming and BrightLink Is Fading Out).

Chiaro says it has raised $210 million to date and employs 221 people. The company has been granted about 25 patents and has filed for 26 more. The company closed its last round of funding, a Series D round of about $80 million, in February. "With any luck at all, we won't have to raise more money," Lewis says.

— Phil Harvey, Senior Editor, Light Reading
www.lightreading.com
Pierre 12/4/2012 | 9:17:46 PM
re: Chiaro Girds 'Router' for the Grid All,

I read an article over the weekend that alluded to further cuts at Hyperchip. Can anyone substantiate these claims, and if so, what departments were affected?

Regards,

Pierre
mr0carrier 12/4/2012 | 9:18:13 PM
re: Chiaro Girds 'Router' for the Grid Are you ready for the results? Lightreading is very jealous since I am beating them to their next router test. Perhaps they will ask me to help them on the next one. On the other hand, maybe not since I seemed to have killed the chat for this article.

For my protection switching test I decided to take the advice of the other poster and see if these bad boys can handle a little dog action, just in case I install one of these in my house. At first I thought I would throw some water on them, but I realized it might take down too much. Instead I decided to use a syringe and inject water into a crack above the route processors. The Chiaro router made a huge blast and a bunch of smoke, much more than the Juniper box. Not surprising since Chiaro was really focused on processing performance and probably jam-packed a mini-super computer into it. Who held up better? Well I have to admit the Chiaro box had a little less impact on the network than the Juniper box. The Chiaro CLI was unthawed, but I had to re-login into the Juniper box. I thought this would be major plus for the Chiaro router but my manager said he doesnGÇÖt care about that since customers donGÇÖt see the CLI and he still plans to double up routers for protectionGǪok, whatever. Monday I get to turn the broken boards back into the vendors, and tell them they just broke. This is the favorite part for the purchasing guys for the reason that they will give the vendors a hard time about disgustingly inadequate faulty equipment, and make the vendors embellish them with lavish entertainment and apologies.

Now on to the blocking test. Here we play with cramming a gang of packets into one pipe. I thought that Chiaro would do better here but they only beat Juni under a special case. I think once I get the Juni TX prototype and start packing packets across a multihop T640 with a TX in the middle that the Chiaro router will start to shine. Reason being, the Chiaro has just one big optical switch (phase array thing-a-ma-bob), however, the Juni box will have to result to a multistage switch configuration which will probably utilize a combination of space and time multiplexing. This means blocking under certain cases. If you are more curious about this there is a minimum pathset method for calculating blocking probability on multistage systems, which is a little easier than Lee Graphs or Jacobaeus methods which are ok, but stick you with those nasty equations you have to program a computer to guess at since you canGÇÖt solve them. Link: http://www.bupt.edu.cn/quick/l...

In summary, the Chiaro router is a cool new techno-toy, but I need to keep my Juni boxes. (next post from mastermind readersGǪGǥwe know who you work forGǥ)

My real boss busted me, he thinks it is funny but is paranoid so:
Disclaimer: This was a virtual router test based on BS press releases, ignorant guestimation based journalism, and article chat noise. It would have been based on some white papers but I am too lazy. Remember boys and girls, Internet chat is a toy not a resource.

It is supposed to stop raining by tomorrow so farewell and have a stupendous weekend, Mr0Carrier
mr0carrier 12/4/2012 | 9:18:14 PM
re: Chiaro Girds 'Router' for the Grid >You must be in Cal Tech or UCSD?

Good guess dude! But, no cigar for you. In point of fact, I am at CSUSM which is handily flanked by them. That allowed me to splice into the new super grid they are building. They have been able to detect the intrusion but they probably figure no one at CSUSM would be smart enough to do such a thing. I have sensed them probing and scrutinizing each other. Of course I needed the manuals. Juniper was easy but Chiaro was much more difficult. It took some doing but I was able to find a backdoor website which gave me access to their internal website. The website is camouflaged, but if you are smart you can figure it out. It is located at: http://www.uans.com/chiaro.htm

Want another pointer about who I am? I might be observed at one or more of the following Starbucks

2183 Vista Way
31 Rio Robles East
411 Cedar St.
332 W. El Camino Real

Cherio,

Mr0Carrier
qqq 12/4/2012 | 9:18:20 PM
re: Chiaro Girds 'Router' for the Grid You must be in Cal Tech or UCSD?
mr0carrier 12/4/2012 | 9:18:35 PM
re: Chiaro Girds 'Router' for the Grid Yo surfer dudes, I finally got a couple of T640s, racked up and powered on. For those of you like GÇ£alloeoGÇ¥ who believe that I have 7ft racks, 20Kwatts of -48V power, and a bunch of test equipment in my house (wink, wink), my wife made me put this high power sucker in my basement on account it is even louder and hotter. I doubt the Chiaro optical switch will save much money but apparently, the lack of a bunch of switch ASICs saves them some power. It also seems to save them a little space. Since the Gibson uses distributed switch architecture each box has to have full blown switch capability.

So the Chiaro box is some new gnarly cool stuff, BUT does it route? Short answer, sorta of. Everyone knows they were lacking Tony Lee the router stud, and my initial test results confirm this. It is no surprise that Juniper rocks protocol implementation and it is no wonder they are being used to route Ipv6 on Internet2 GigaPops. Which by the way I am glad the Uni-Profs are playing with Ipv6 before I have to. I am sure it will make my test plans blossom into bushes. Juniper is overall winner on this round. Chiaro did beat them out on table cap, and LML. That did not surprise me, since they do have supercomputer expertise. I had noticed ahead of time that the CLI ran way faster on the Juniper box. Now donGÇÖt jump on the Juniper bandwagon yet though, still got to look at protection and blocking. Here be the router scores (higher better):

IP tests: Chiaro-4 Juniper-5
MPLS tests: Chiaro-2 Juniper-4
Longest-match lookup: Chiaro-4 Juniper-3
BGP table capacity: Chiaro-4 Juniper-3
MPLS LSP capacity: Chiaro-2 Juniper-4
Route flapping: Chiaro-3 Juniper-3
Convergence: Chiaro- 2 Juniper-3
Filtering: Chiaro- 3 Juniper-3
Class of service: Chiaro- 2 Juniper-4

Cheers,

Mr0Carrier
arch_dude 12/4/2012 | 9:18:45 PM
re: Chiaro Girds 'Router' for the Grid Bezelbutt said:
"Juniper is OEO, Chiaro is OEOEO. The Chiaro router is thus bigger by far, to accomodate all these optical interfaces going to the fabric.
(refering to M-class Junipers)"

Basically a T640 or other scalable router is a switching fabric surrounded by linecards. There are two major classes of switching fabric: multi-stage and scheduled crossbar. The Juniper T640, Caspian, Pluris, and Hyoperchip use multistage. The switch-fabric vendors (PMC, Agere, Zagros, etc.) mostly sell scheduled crossbar fabrics.

A scalable router built around a scheduled crossbar generally connects the linecards to the fabric via cheap VCSELs, so the router is OEoEoEO
(using little o for cheap optics.) The crossbar is electrical.

Chiaro also uses a scheduled crossbar, but they have a neat new trick: the crossbar switches photons, not electrons, so they are OExEO where the x switches photons. Note that the scheduler
(the really nasty part of the scheduled crossbar) is still electronic, not photonic.

I got all this from a careful reading of the story on EET.
skeptic 12/4/2012 | 9:18:46 PM
re: Chiaro Girds 'Router' for the Grid HFR (cisco)
T640 (Juniper)
Caspian


What makes anyone chase Caspian? What share of the core router market do they have?
-------------------------

By "chasing" what I meant was competing hard
against. The HFR/T640/Caspian/Chiaro will
compete for any business against each other.
All four are new products with only the T640
possibly having any market share.



Belzebutt 12/4/2012 | 9:18:47 PM
re: Chiaro Girds 'Router' for the Grid each PFE provides the equivalent of 2xOC-192/STM-64 interfaces to the network
...
I can't see why Chiaro's router should be different.


The difference is, Juniper provides the *equivalent* of such and such speed towards the fabric. Chiaro actually has physical interfaces of such and such speed towards the fabric.

Juniper is OEO, Chiaro is OEOEO. The Chiaro router is thus bigger by far, to accomodate all these optical interfaces going to the fabric.

I'm not comparing this to a Juniper T-Matrix however (or whatever they call a bunch of T-640s connected together), if you compare the Chiaro to that it becomes more reasonable.

But like I asked, service providers need a Chiaro router just about as much as they need a Hyperchip router or any of the other Petabit routers that have gone belly-up.
Belzebutt 12/4/2012 | 9:18:47 PM
re: Chiaro Girds 'Router' for the Grid Try:
HFR (cisco)
T640 (Juniper)
Caspian


What makes anyone chase Caspian? What share of the core router market do they have?
skeptic 12/4/2012 | 9:19:02 PM
re: Chiaro Girds 'Router' for the Grid uhm, great so they're chasing who pluris?
----------
Try:
HFR (cisco)
T640 (Juniper)
Caspian

The people who are chasing them are:

Alcatel
Hyperchip


Bluebeam 12/4/2012 | 9:19:03 PM
re: Chiaro Girds 'Router' for the Grid Belzebutt wrote:

A current router with one OC-192 and 15 OC-48 ports needs only one OC-192 interface and 15 OC-48 interfaces.
This new type of router would need all of the above, plus 16 more 10Gig interfaces


Unless of course you have four OC-48 ports per card, which access the fabric through the same ingress packet processor. Or your line rate through the switch is not 10G but, 622M with 4 or 16 parallel paths through it, or ... Again, this is not different from any fabric switch based today.

Take Juniper's T640, for instance. From a white paper "each PFE provides the equivalent of 2xOC-192/STM-64 interfaces to the network and more than two OC-192/ STM-64 interfaces to the switch fabric to ensure high performance and eliminate head-of-line blocking." A PFE is located in a FPC, and each FPC can house multiple ports. I can't see why Chiaro's router should be different.

I'm not discussing the viability or economics of their approach, just pointing out that mixed interfaces are not necessarily a problem.

Bb
BobbyMax 12/4/2012 | 9:19:04 PM
re: Chiaro Girds 'Router' for the Grid There is not much new here. Previosly swiching and routing functions have been combined into one. Chiaro combines the switching abd rouing functions into a single box.

It is not clear if commercial deployment of Chiaro technology would ever occur.
mr0carrier 12/4/2012 | 9:19:04 PM
re: Chiaro Girds 'Router' for the Grid >Oh, I thought you were a carrier/telecom guy... >now you are an IT guy huh?

The point is everyone at a carrier, service provider, telco, or even an IT guy not working yet would have a clue about layer 2 and 3 differences.

>Or are you a professor at UCSD?

Think I'm a prof? Cool!

Cheers,

Prof0Carrier
alloeo 12/4/2012 | 9:19:04 PM
re: Chiaro Girds 'Router' for the Grid >I thought they were lying about being able to handle 100GÇÖs of OC192 ports, but with an optical switch at the core I guess they are probably just starting.

Please enlighten us, where are your 100's of OC192 line cards? or does USCD even have a handful of OC192 cards at all?

>I would hope a new grad IT guy would realize you donGÇÖt want to link separate University campuses together through a freakin layer 2 switch

Oh, I thought you were a carrier/telecom guy... now you are an IT guy huh?

>my son... my wife...

Mr0Carrier, do you really work for a carrier? Or are you a professor at UCSD? Obviously, from what you wrote, there is no diffrentiation between 'carrier' and 'service provider' or telecom and IT. With all due respect, does the university really allow you to play with such an expensive quipment at your house? what if you dog comes and $(*% on it?
mr0carrier 12/4/2012 | 9:19:05 PM
re: Chiaro Girds 'Router' for the Grid I am testing the Chiaro router now. So far I think it is pretty awseme. I thought they were lying about being able to handle 100GÇÖs of OC192 ports, but with an optical switch at the core I guess they are probably just starting. The switch module looks neat, I wish I could take it apart. To bad no one will need their full scalability for a while. I also think the STAR technology is pretty cool. Waiting 20 minutes for a Cisco route processor to reboot is irritating.

I donGÇÖt know why someone suggested it was only a switch installed for the optiputer. I would hope a new grad IT guy would realize you donGÇÖt want to link separate University campuses together through a freakin layer 2 switch.

This router is big, physically that is, although nothing like the Avici iron barn that will cave your floor in. Although I am having fun with it, my son is a little irritated with the latency. It interferes with his online gaming. It is not really the routers fault though. Since it does not have an interface smaller than OC12 I had to borrow an ADM to connect our PCs together since we just have GigE on them. My wife is a little irritated with the fan noise, which by the way is raising my electric bill since that router does put out some hot air.

Cheers,

Mr0Carrier
achorale 12/4/2012 | 9:19:06 PM
re: Chiaro Girds 'Router' for the Grid FatherConfessor wrote:

> Wrong! Look at their web site. It is a
> fullblown internet core router with an
> optical backplane"

uhm, great so they're chasing who pluris?

achorale

alloeo 12/4/2012 | 9:19:22 PM
re: Chiaro Girds 'Router' for the Grid Well, from a pure technical/research point of view, what chiaro did was great. they made a big step toward optical sub-wavelength switching and actually is trying to sell a prototype. the europeans, esp. the alcatel people in france have been trying many years and burnt a lot of money and still got no where.this GaAs AWG ns switching technology, is really COOL.

but, if they are trying to sell it as a product, you can't blame the engineers for doing that at the peak of party. it was the stupid investers who threw money at them, and it's the investers who will lose that money. i think the engineers there are doing an awsome job, it just cracks me up to see how the marketing people still try to creat a buzz at this time. sure they are talking so SP's, so was everyone else. sure they sold one unit, but it was bought with NSF money, which should be spent not on conventional in stock products. i hope this was not the only prototype developed that they sold to UCSD.

skeptic 12/4/2012 | 9:19:26 PM
re: Chiaro Girds 'Router' for the Grid Even if the above would work (which I still doubt), are the advantages compeling enough to justify implementation? I'm not sure...
-------------

Two years ago, you would have been able to
raise tens of millions of dollars on that idea.
And its a better idea than many that got funding
at that time. There is no money for it now.

People looked at putting control signals
(packet headers) on a different wavelength so
that the headers could be OEO'ed while the
data was a pure optical signal.

The problems were synchronization, dealing with
locally terminating traffic and slow-speed
switching technologies (MEMS mostly). The
switch speed of MEMS pushed lots of people off
into "burst switching" which was never practical.


lightfantastic 12/4/2012 | 9:19:26 PM
re: Chiaro Girds 'Router' for the Grid So is Chiaro destined to become a footnote in optical history, or this a viable business plan buried in there? They have been to SPs, but is there s/w platform being tested anywhere? Is anyone testing their h/w?

This project is interesting, but as far as I am concerned, that's as far as it goes.

-l
achorale 12/4/2012 | 9:19:27 PM
re: Chiaro Girds 'Router' for the Grid >Wrong! Look at their web site. It is a fullblown >internet core router with an optical backplane"

>There is a huge difference between web pages and >reality.

>Well, if you want reality, they are out talking
>to the large service providers trying to sell
>a core router. They have been privately
>demonstrating a core router for the better part
>of a year and the grid computing people who
>picked the product were out looking for an IP
>router.

Yeah, whatever. Guess we know who you work for
Belzebutt 12/4/2012 | 9:19:28 PM
re: Chiaro Girds 'Router' for the Grid The key I guess is that the line rate towards the backplane doesn't need to match the external line rate. If you think about it, that's not different from an electrical switching fabric.

You're right, in that case the rate towards the backplane needs to be the rate of the highest external port on the router. This is even worse. It is different from today's routers:

A current router with one OC-192 and 15 OC-48 ports needs only one OC-192 interface and 15 OC-48 interfaces.

This new type of router would need all of the above, plus 16 more 10Gig interfaces. One OC-192 to send the one external OC-192 signal towards the optical fabric, and 15 more fabric-facing 10Gig's at each OC-48 line card to read this 10Gig signal, so the router can then do packet manipulation or whatnot, then convert to OC-48.

I have a hard time believing that this architecture can be even close to cost-effective today, since the price of today's routers is mostly in the interface cards. Don't forget that ASICs get faster and faster too, and we are nowhere near saturating today's fastest routers and there's no sign that the demand is there.

fw23 12/4/2012 | 9:19:28 PM
re: Chiaro Girds 'Router' for the Grid >Wrong! Look at their web site. It is a fullblown >internet core router with an optical backplane"

>There is a huge difference between web pages and >reality.

Well, if you want reality, they are out talking
to the large service providers trying to sell
a core router. They have been privately
demonstrating a core router for the better part
of a year and the grid computing people who
picked the product were out looking for an IP
router.

achorale 12/4/2012 | 9:19:29 PM
re: Chiaro Girds 'Router' for the Grid Bluebeam wrote

"The key I guess is that the line rate towards the backplane doesn't need to match the external line rate. If you think about it, that's not different from an electrical switching fabric"

Then what happens to the remaining 3 OC-48's worth of data? Even the biggest routes perform aggregation and re-grooming since even the biggest pipes consists of thousands od smaller information streaams. Unless there is an electrical switch hidden somewhere in there, they are not able to re-map these streams. It's simply aggregated OC-x in and aggregated OC-x out, without looking at what's in the aggregated stream. Which makes it a switch.
skeptic 12/4/2012 | 9:19:29 PM
re: Chiaro Girds 'Router' for the Grid
Its not the "all-optical router" but the switching
technology could eventually serve as the basis
for one.

There were/are several different problems that
need to be solved to get to true OOO. Their
switching approach, as described, is more pratical
for reaching OOO than anything else currently
out there.

If you could solve the cell marking problem and
the optical buffering problem, you have reached
OOO.

We are still a long way from true OOO packet
switching, but getting something like this working
was a milestone on the road to OOO that needed
to be reached.





achorale 12/4/2012 | 9:19:30 PM
re: Chiaro Girds 'Router' for the Grid FatherConfessor wrote:
"Wrong! Look at their web site. It is a fullblown internet core router with an optical backplane"

There is a huge difference between web pages and reality.
Bluebeam 12/4/2012 | 9:19:34 PM
re: Chiaro Girds 'Router' for the Grid Belzebutt wrote:

And if the fabric is all-optical and bit-rate agnostic, ask yourself, what happens if your router wants to switch a packet from an OC-192 interface to an OC-48 interface? One fabric-facing interface will be and OC-192, optically switched to an OE OC-48 interface... it won't work!


The key I guess is that the line rate towards the backplane doesn't need to match the external line rate. If you think about it, that's not different from an electrical switching fabric.

Bb
boozoo 12/4/2012 | 9:19:36 PM
re: Chiaro Girds 'Router' for the Grid The problem with optical packet switching is that
because the routing information is in the packet itself, you can't switch it until you look at the packet. At that point, it's too late, because part of the packet is already gone.
So if you are looking for a magic solution to route IP packets in the optical domain, look for a solution to buffer/stop the light.

Here's an idea (coarse) (which may or may not work) to do a "sort" of hybrid packet switching:

- On the ingress card, optically tap out the packet header and convert it to electrical so you can do the header processing in the electrical domain. At this point you know where to direct THE REST of the packet, so you can drive the switch to send it to the correct destination. After the packet is finished, insert the header.

- On the egress card, receive the packet followed by the header, buffer if electronically (you can't avoid this) and create the outgoing packet.

The advantage is that you don't have to store entire packets on the ingress datapath. All you need is a fast NPU that would do a fast analisys of the packet header and drive the switch in time so that the payload is directed to the right egress port.

The NPU will also have to hunt for the "end of packet".

Even if the above would work (which I still doubt), are the advantages compeling enough to justify implementation? I'm not sure...

Boozoo

Belzebutt 12/4/2012 | 9:19:37 PM
re: Chiaro Girds 'Router' for the Grid The real killer is the line card... the cost of the whole box is usually dominated by the linecards, esp. at OC192 and one hundred waves per fiber.

This is a good point, and another thing I wanted to add to my post.

If this router does OE conversion at the line cards in order to do IP lookup, and then switches back to optical to send packets to the optical switch fabric, then it needs 2x as many OE interfaces as a conventional router: one line-facing OE interface, and one fabric-facing OE interface. These fabric-facing optics are probably cheaper, but add cost nonetheless.

And if the fabric is all-optical and bit-rate agnostic, ask yourself, what happens if your router wants to switch a packet from an OC-192 interface to an OC-48 interface? One fabric-facing interface will be and OC-192, optically switched to an OE OC-48 interface... it won't work!

Unless I make totally wrong assumptions about this "router", it has some serious limitations. So either:
- It's only a switch, not an IP router (not likely judging by their website)
- They do OEOEO, not OOO, and can only optically switch between ports of the same speed
- They do some kind of magical optical IP packet lookup and they can switch between different speeds of signals without converting to electrical (in which case LR should make a much bigger deal about them)


So LR, how about clearing up this little mystery for us? At least some good old-fashioned speculation is in order.

alloeo 12/4/2012 | 9:19:38 PM
re: Chiaro Girds 'Router' for the Grid >So if the bottleneck for router vendors is how high their switch fabrics can scale, then they solved that bottleneck.

I think the conventional electrical switch fabric will reach its limit when you simply can't squeeze enough swich chips onto one single switch card and the signal traces are just overwhelmed on the backplane. But even up to that point, the physical causes (heat, space, etc.) are the limiting factor, not the cost.

The real killer is the line card... the cost of the whole box is usually dominated by the linecards, esp. at OC192 and one hundred waves per fiber. As long as they still use OEO line card, they simply can't slash their cost that far.

Software is totally another issue. I don't think Chairo for its size can afford a full-blown software team. Mostlikely they will buy someone's code.

Weren't they sued by Alcatel before?
Belzebutt 12/4/2012 | 9:19:38 PM
re: Chiaro Girds 'Router' for the Grid I would guess that they send optical cells across the switch. You could schedule the cells to
be send across the switch without looking at
the cells (I think).


I think this poster has it right.

There's no way they can look up IP packets optically, they have to convert them to electrical, do a look-up, and then convert them back to optical and switch them. It appears that their big breakthrough is replacing an electrical switch fabric with an optical switch fabric, like a Corvis OOO switch that can switch extremely fast.

So if the bottleneck for router vendors is how high their switch fabrics can scale, then they solved that bottleneck.

If the capacity of switch fabrics in today's routers is not the main problem, then this is basically a cool gadget. But what about routing protocols? This thing still needs excellent routing protocol support and interoperability in order to be a contender.

Like we saw in Hyperchip's case, simply having huge capacity means nothing today. Very few people need T640's, who's going to need this monster?

If there is a lot of value in this switch fabric, I would say these guys will get bought out by an established router vendor.
boptic2002 12/4/2012 | 9:19:41 PM
re: Chiaro Girds 'Router' for the Grid >Anyway, the voltage controls the location of the >constructive interference, which in turn creates >bright spots that allow the light to be steered >and aimed at any one of the 64 outputs on the >switch module and -- voila! -- the light is >switched in nanoseconds by a machine with no >moving parts.

This just sounds [email protected]!%ing cool.

light-headed 12/4/2012 | 9:19:43 PM
re: Chiaro Girds 'Router' for the Grid let's do what a conventional router does at a much higher cost. no one needs a scalable backplane based on optics today.
skeptic 12/4/2012 | 9:19:45 PM
re: Chiaro Girds 'Router' for the Grid There was some internal debate about this. This article focuses on the equipment that was actually shipped as part of grid-computing project. The way that equipment was described, it's more of an optical switch.
--------------------------

Based on talking to people, I dont think thats
right. I think they are using an IP router
as a giant interconnection hub. If you put
enough computers/small routers together with
high-speed interfaces, you end up with a router
network.

And if you think about where the grid computing
people are going, they would want a large
router eventually for the wide-area high
bandwidth grids that some of them talk about.

I'm not sure what it would not be a router.
I can't see the grid people using a propritary
interconnect. They all want IP these days.





alloeo 12/4/2012 | 9:19:45 PM
re: Chiaro Girds 'Router' for the Grid >I dont think this is correct. Chiaro is building
>a router with an optical switch. It is a router.

If I understand correctly from what you said, fw23, what they are building is ~NOT~ an optical packet switch in the strict sense where the payload goes through no OEO conversion and get switched all-optically. What they are building is an IP router with an optical back plane that got rid of all the limitations of conventional electrical backplanes. If this in deed is the case, then all the packets still under go OEO conversion, electrical header processing, electrical buffering/scheduling, etc., just like in the big ole routers, it's just the signal path from one line card to another (and the reconfiguration of inter-connectivity) is optical.


>In any case, they are out trying to sell a
>core router to the big service providers.

What will be their edge over Cisco or Juniper or other guys who are doing the same thing (take Stanford for example)? The cost of GaAS AWG's is not cheap (BTW do they fab their AWG's in house, or they buy from someone?), let alone the control circuit to do ns scale switching.

It will be interesting to know their box's price and if it does all the other stuff a true IP router does.

Scott Raynovich 12/4/2012 | 9:19:47 PM
re: Chiaro Girds 'Router' for the Grid There was some internal debate about this. This article focuses on the equipment that was actually shipped as part of grid-computing project. The way that equipment was described, it's more of an optical switch.

Chiaro is promising more "routing" functionality in the future (including they stuff people are talking about on the Web site). But that was not in this prototype.
FatherConfessor 12/4/2012 | 9:19:49 PM
re: Chiaro Girds 'Router' for the Grid > This is a wavelength switch not an optical
> packet switch.

Wrong! Look at their web site. It is a fullblown internet core router with an optical backplane.

fw23 12/4/2012 | 9:19:49 PM
re: Chiaro Girds 'Router' for the Grid >Can someone explain how this switch provides >packet
>switching functionality?

The line cards are OEO. Their optical stuff
can switch at high enough speeds that they
can do packets.

Various people tried to do the same thing with
MEMS over the last couple years, but the switch
rate of MEMS was so low that packet switching
didn't work. Various people tried schemes like
burst switching, but none of them ever went
anywhere.





>The explanation in the article
>does not indicate any mechanism to look at what the
>light wave actually carries. Perhaps the control
>engine which steers the light does look at it.
>Can someone shed some "light" here?

I would guess that they send optical cells across
the switch. You could schedule the cells to
be send across the switch without looking at
the cells (I think).


DoTheMath 12/4/2012 | 9:19:50 PM
re: Chiaro Girds 'Router' for the Grid Can someone explain how this switch provides packet
switching functionality? The explanation in the article
does not indicate any mechanism to look at what the
light wave actually carries. Perhaps the control
engine which steers the light does look at it.
Can someone shed some "light" here?
achorale 12/4/2012 | 9:19:50 PM
re: Chiaro Girds 'Router' for the Grid This is a wavelength switch not an optical packet switch.



fw23 12/4/2012 | 9:19:51 PM
re: Chiaro Girds 'Router' for the Grid >Chiaro's initial prototype isn't really >a "router" per se, but rather a giant, super->
>fast optical packet switch.

I dont think this is correct. Chiaro is building
a router with an optical switch. It is a router.

In any case, they are out trying to sell a
core router to the big service providers.



HOME
Sign In
SEARCH
CLOSE
MORE
CLOSE