Is Hyperchip Hype?

Peter Galbraith writes "There was an interview on CBC (here in Canada) last evening about Hyperchip, a Montreal-based company that are working on a new type of router that would scale up 1000 times in traffic (so wouldn't be obsolete in less than a year) and would pass packets to their destination in a few hops instead of a dozen or more. Any experts out there think it's hype? Or real?" The explanation on Hyperchip's "technology" page is pretty thin, but considering they just raised $70 million, I hope they've given more convincing details to their investors.

Comments

Step One:

[Saint Aardvark]

Step One: Tell everyone you have an amazing new router.
Step Two:
Step Three: Profits!

Re:Step One:

[ekrout]

Step Two: Sign-up for a user account at the ultimate geek site and post a story about Step One.

;-)

Funny Coincidence

[Zenithal]

I'm here in Montreal, and I applied to go work for them not even a week ago.

I was inadequate :)

If they can figure that out, they probably have a chance.

Hyperchip

[yoink!]

I was at a Pub one evening (I live in Montreal) and I happened to meet their sales manager... ms. Jen Goldfinch. Although I had seen the Hyperchip building on many occasions, I had never inquired as to what they do. After meeting this woman, I was given the impression that their routers are actually in use by some of the big players in the digital pipelines game. She was actually pretty clear on that, although I can't seem to find any exact information concerning their customers on their website. Perhaps some questions to nortel, and qwest folks might clear this up. The only thing that make me dubious about her claim of widespread adoption, would be that if their products are so much "better" (for the lack of a better word) than the competitions, then why is abilene using cisco products? Unfortunately I don't have that kind of time on my hands.

Hopcount != Speed

[cwsnate]

Simple hop count does translate directly into speed. A two hop route plagued with slow nodes might be (and often is) slower than a route that consists of more hops with faster nodes. Sounds like ignorant marketing hype.

Other vendors that come close?

[scubacuda]

Cisco and Juniper can only (currently) route in gigabit speeds.

Other competitors that they will have to deal with: Pluris, IronBridge Networks and Charlotte's Web Networks .

Use Existing Technology

[Anonymous Coward]

Cisco has had the technology to do their so called 'smart routing' for years. It's called BGP. I have seen it intelligently route packets across the country that only goes across 2, maybe 3 routers. It seems in this "dot-com" era of techies, no one has a clue about routing. There are so very few people who know what to do with a router, and obviously those people aren't working for any commercial ISPs. I used to work networking for the government, and one of our customers had a problem with routing through UUnet. I contacted their router dept and talked to one of their so-called "router engineers" and had to explain to him how to fix an assymetrical route.

Do a traceroute to yahoo.com. Conceivably, you shouldn't have to go more than 5 hops. But with every major corporation creating a string of routers rather than a mesh.... it takes for ever to get there.

Re:Use Existing Technology

[juuri]

Open foot, insert mouth.

CISCO actually preaches in their advanced networking design (one of the things needed to get a CCIE) that all intelligence should be moved outside the core. The core exists only to switch they scream time and time again. They are right.

Examing packets is damn expensive, you don't want that in your core layer at all. You want it moved out as far as possible to move the possible bottleneck as far towards the end user as possible. A well designed network does its job well without having any fricken idea whats going on other at any high levels.

Conversely the upside of this belief for vendors is it helps to sell more equipment since you need more layers to properly shield the core from having to examing the packets.

Re:Use Existing Technology

[Phasedshift]

Couple of things...

1. 90% of the time when you call a large internet service provider, you speak to their frontline support, who address their level 2 support (etc) as 'routing engineers', consider how many people must call them and complain about problems (90% of which are probably caused by stupid things like them advertising a /24 in class B space, without advertising the /16 also, so the route gets nuked at a border router, as quite a few providers filter based on classfull boundries, or people just not understanding how a traceroute even works, and demanding to speak to a 'routing guru' because their traceroute dies after a certain point due to an ACL (access control list), so they naturally think the web server/mail server, etc they are going too is down, even though it is not.).

2. As someone pointed out, number of hops != latency, etc. Most people who are just starting their quest for knowledge in this field tend to confuse the two unfortunately.

Now, while in an ideal network, 99% of things will be done at layer 2, thereby making the total hopcount in your traceroute lower (if your traffic is going through an ATM switch who doesn't know about/care about layer 3 information in an ATM cell, it will obviously not show up as a hop in your trace). The hopcount your traceroute shows doesn't matter, the simple fact that all that has to be done is the header of the ATM cell (the first 5 bytes) is read, and then the cell is forwarded to its proper destination (similar for ethernet, using cut-through switching, etc) doing switching at layer 2, vs. having to read the IP packet's headers to find the destination will provide a noticible decrease in latency in most situations.

You are however correct in that Cisco has many features out there that will greatly increase performance for people. Mind you, there aren't any official 3rd party benchmarks against this (the company the article is referring too) company's products, so we don't know if they are something to laugh at, or really are better then whats out there currently (although I am voting for the laughing part).

Also, in regards to BGP, (heh btw it has quite a bit of overhead, since it uses TCP), while it is pretty much the only good choice for an EGP (external gateway protocol), you will still need an IGP such as OSPF, etc unless you intend to have your router(s) have BGP sessions with the IP's of other routers known via directly connected (or static routes), which would be stupid (except for some situations using static routes in a very small network). Its not like you setup a BGP session with another router, and it 'magically' works, there is quite a bit of traffic engineering involved (how much is dependent on how big your network is), and cooperation among internet service providers (i.e. to set the localpref that is distributed to your IBGP mesh based on certain communities received from a peer, or the other way around, so people can control the path traffic goes back into their network in a better way then padding the route (i.e. adding more AS's to the AS_PATH which chances are wont give you the desired results), or other methods).

Re:Use Existing Technology

[Bookwyrm]

Anonymous Coward wrote:

Building it is one thing. Making it run efficiently is something else altogether. Perhaps that takes a class of people even smarter than the builders?

That is what I was referring to in my comment in a sense. Anyone can build chaos, diversity -- a jungle. Anyone can say "We will make the edges smart and the core dumb!", but the doing is another matter. Such a construction is not efficient, though. History tends to indicate that farming has won out over jungles, that civilization over barbarians, corporations over small buisinesses.

As the demands placed on the network become more stringent in terms of QoS, costs, efficiency -- once the network becomes no longer cutting edge, but boring old infrastructure, it will become ordered and managed to increase efficiency at the cost of diversity. Every frontier is eventually settled because it uses resources more efficiently that way.

Pretending that the core network does not need intelligence allows people to pretend that they can do whatever they want, drop whatever packets they want into the system, and have it work. (In the slightly higher application level, people learned that they could send SPAM to everyone else easily.) The issue is that this is slowly creating an hostile relationship with the core network. Issues with having ports blocked to prevent people from running servers on their connections are symptomatic of this -- the edge and the core are conflicting over what is allowed. (Hint: the core wins unless the edge pays at least the cost of the actual bandwidth used. The core can always cut the connection. If the edges do not cover the costs of the bandwidth, the core goes out of business, and everyone loses (you know, like all those DSL providers...))

Barbarians who have stood at the edges and shouted and screamed that they do not need civilization have always been pushed back by the expansion of civilization. The edges will have to be come civilized and deal with the core network in a civilized fashion, not just run rampant through it -- the core is alread setting up walls and gates on the edges to stop the barbarians: filtering out spoofed source addresses, NAT, firewalls, port blocking, monitors, sniffers, etc. Because the core network is denied any useful information/intelligence about services to make optimizations, because it has to handle *anything*, which is inefficient/expensive, the core network is , unsurprizingly, becoming dumb -- and, surprise, dumb things do not do well in performing a variety of services efficiently, the services and traffic have to be simple and uniform to be handled efficiently by a dumb box. So, the traffic and services are *made* simple -- no web servers for you, bud! We'll forcibly shape your traffic patterns until they are more simple and efficient for us, 'cause we're only a bunch of dumb core guys, right? We can't do anything complex! Why, yes, tech support really is staffed by low-grade morons -- the network is pretty dumb and simple, you know, doesn't take any brains to run!

Ultimately, the barbarians will have to become civilized, the only question is on whose terms, or find a new frontier. If the barbarians are going to want good terms, they are going to have to treat the core with respect -- they will have to have intelligence in the core to communicate with it. For QoS, etc., the edges will have to be able to tell the core what they want, and the core will have to have the intelligence to see if it can manage it, and arrange for the service. And the barbarians better negotiate while they still have power -- if they all get crowded into little reservations of network places with unfiltered services and become a tiny part of the market first without any bargaining power, well...

If the only communication/negotiation between the edge to the core is "Route packet! Route packet! Route packet!" The only reponse the core can have is "Yes" or "No". Allowing for more intelligent negotiation of service allows for compromises and more flexibility. The core can *always* say no.

Murphy's Golden Rule: Those who have the gold make the rules.

Possession is 9/10ths of the law.

As long as we need core networks, routers, interconnect points, etc. those who own them will ultimately have the power to make the rules. And they aren't owned by the IETF, and they aren't owned by the edges.

Could be...

[Atomic Frog]

We had a project that was probably similar in concept, maybe slightly lower throughput. I don't think the claims are that far fetched.

However, getting one chip working is one thing. Getting an entire box is a whole 'nother trick entirely, as I'm sure they will discover.

They are, obviously short on technical details, but I find no particular reason to disbelieve them. There are a lot of "real" tech companies in Montreal (my ex-company had a branch there), and fewer fakes than other places.

$70million won't last you very long without any other source of revenue. If they are lucky and really, really good, they may have a product out in production in 2-3 years.

The terabit market flopped, so go faster!

[Dave Goldblatt]

This is a company on its fourth round of financing ($220M CDN invested to date), with no announced customers - or even beta trials.. and they've been around since 1997?

Another thing is that article is misleading; they really received $12M in funding, and added another $31M in repayable loans from the Canadian government. Again, the numbers quoted in the article are Canadian dollars, not US.

Several terabit router companies have failed (such as Ironbridge ) and others are having problems, a la Avici along with Nexabit.

For more entertainment, read the article and comments in Light Reading.

It's not the bandwidth, it's the services. Besides, who can afford to provision 65,000 OC-192s?

patents

[Syre]

they claim to have 41 patents issued... I found 3:


I/O and memory bus system for DFPS and units with two or multi-dimensional programmable cell architectures

Efficient direct replacement cell fault tolerant architecture

Fault tolerant data processing system fabricated on a monolithic substrate

From these it appears they are fabricating wafers with lots of semi-independent processing nodes, which are tolerant of failures of some of the nodes (and can therefore take into account chip production glitches on part of the wafer).

This could give them a potentially large performance advantage, if they can do it right.

More info

[Alien54]

There is Investor analysis here, with a power point presentation

They also have an EETimes story Archived and there is this news item from before the dot-com boom went bust.

Other items include this bit saying we don't need petabit routing anyhow (just wait a few years!). I also spotted this job description from some namesless company.

System Engineering Manager As the System Engineering Manager, you will be responsible for the Petabit/Terabit router prototype system development using the state of the art switch fabrics. You will lead the system design team to perform OC-768, OC-192, OC-48 linecard design, and multi-Terabit switch plane design. Requirements include a BSEE plus 7 years chip design experience or MSEE plus 5 years design experience in router system and high speed board design. Networking and Gigabit line card experience is preferred.

Basically, this job description says to me, "You will invent the products we need so that we can make lots of bucks off your brains". One of those things, go in with eyes open.

Same thing, other side of the border

[nsample]

The same thing is happening on this side of the border. LaurelNetworks is another super-startup company, with a lot of capital, with their guns aimed at the "big boys" of network hardware (Cisco, Juniper, etc.). I did some work with Laurel, and even some work with a micro-startup networking company, BlueWave Networks.

I don't think that there's any hoaxing at all going on here. They're legitimate players with some heavy capital backing them. They also have great engineers and some good technology. It may not be enough, however. What it's going to come down to (IMHO) is the willingness of big ISPs and carriers to adopt technology from a new vendor.

Cisco may not have the best equipment, but everyone and their dog worth their salt in this game knows IOS and how to admin it. You can't say the same of any of the new vendor's products.

We've moved beyond the days of "great ideas" and "great products." Internet routing is a mature market in which the biggest obstacle is now overcome the inertia of the entrenched players.

The anology reminds me of Linux vs. MS, but then again, what doesn't? :)

Destination-based Wavelengths

[vtechpilot]

From Thier site:

Accelerating the war are the recent advances in ultra-long-haul optics and optical switches, which are making it practical for routers to place packets on destination-based wavelengths that can take them as close as possible to their final destination in the core, thus eliminating intermediate routing hops and unnecessary O-E-O conversions.

I read this as meaning that you have a strand of fiber that runs from say for example Boston, to NY to Washington to Richmond to Charlotte to Atlanta to .... and that a packet originates in Boston and is transmitted in red light to the next node, if the light isn't the color for that node, So if red was Richmond, NY and Washington would reflect the red light down the line until it gets to Richmond. This way NY and WAS don't have to convert the signal which saves a lot of time and work. Also this allows Boston to talk to Richmond as 1 hop as far as the software is concerned. Similarly a packet transmitted in Amber light might go all the way to Atlanta before being converted to Electrons.

Sounds like a good idea to me since it should work on existing fiber. The real question is how hard and expensive is it to start with two nodes and grow from there.

Moore's Law

[Apreche]

Look at the graph on the company's white papers. Optical vs. Moore's law. First of all you really can't compare a law and optical. Second of all, they have moores law wrong.

It's the expandability of their backplane....

[addikt10]

That makes their product different. With Cisco gear, you have a very, very small number of high speed ports available to you in one chassis (and several of these "capabilities" are eaten up when you go with redundant solutions).

The idea behind Hyperchip, that is a supposedly better implementation than Avichi's that preceeded it, is to have a packet switching backplane that is expandable to multiple bays, as opposed to the tiny boxes such as the cisco 12000. Since it is a common backplane, there are fewer "hops".

The real limit is power. The Avici systems used over 400 amps of three phase power per bay, and (I believe) scaled to 16 bays, each one capable of running over 60 OC-192s at line speed.

Hyperchip's unit looks better.

Right now, their just trying to figure out how to market it, and how they can add services inside the box that you wouldn't get otherwise. Think of all the stuff you would like to do to streams of that size, but just can't. Also, think of what to do with packets that are going to full pipes. At OC-192 speeds, you can't hold on to packets. There isn't enough time to put it in to memory.

PS OC-192 can carry approximately 10 Gb/s. Or, over 1.2GB/s (this isn't ethernet, it's sonet) (but will have 10 Gb ethernet interfaces, but they just can't carry as much)

At 10 Gb/s, a 1600 byte packet (for those ethernet fans out there) is on the wire (going across a fixed point)for 160 nanoseconds.

Lucent used to have a router that sounds like this

[thogard]

But the never found any customers. It turns out that the few people that need very high speed routing don't buy anything but cisco and in that market it won't matter if you have a product thats much better, you'll have poor sales for years before the market will even consider your product.

As far as routing much faster, its not that hard to do. If you stop treating a router as a router and more like a switch, you can speed things up a grat deal with content addressable memory (the stuff used for cache tags). Its very expensive but 8 mb of CAM ram will let you decide which of 16 interfaces a packet goes to within 500 ps after the address bits hit the hardware. You can't do real time route update on this type of system like a cisco but you can still change routes within miliseconds.

The ideas behind the internet are dead when a small business can't dual home. Without routable class C address, that has already happened.

Check out Terabitcorp

[Broadcatch]

At TerabitCorp, Alan Huang employed what he calls a Galois Network using many standard off-the-shelf routers to create a fault-tolerant, open platform terabit meta-router with some very cool properties. Not sure if the idea ever flew, though...