Hyperchip Lands $43M in Financing

Hyperchip Inc. said Friday that it has secured another US$12.4 million in venture funding and about $31 million in repayable loans from the government of Quebec. This brings the company's total funding to date to about US$136 million.

The company says its router, designed to handle up to 65,000 OC192 (10 Gbit/s) interfaces per system, is in trials with one large carrier (see Hyperchip Hypes Its Hardware and Hyperchip Raises $67 Million ).

"We've completed our first beta trial and were pleased with the results," says Hyperchip CEO Brian Barry. "Now we're hardening the product -- testing and retesting it to make absolutely sure it is what it should be."

The company is planning additional trials with an unnamed carrier and will release its product later this year.

Hyperchip's funding round, its fourth, was led by TechnoCap with investments from Vertex Technology Fund, Advent International, Optical Capital Group, JT Ventures, Artémis, and Pilgrim Baxter.

The company has grown to 280 employees so far, several having joined recently, when IP Optical went under (see IPOptical Fades, Hyperchip Feasts and Hyperchip Hires Hershey).

— Phil Harvey, Senior Editor, Light Reading
<<   <   Page 8 / 8
speed of light 12/4/2012 | 11:02:53 PM
re: Hyperchip Lands $43M in Financing Guys,

Sorry to bother this discussion with a different subject, but I just received an invitation from Lightreading, about the launch of "inside Light", dedicated to Marketing ans Sales in the Telecom field. Well I think it's a pretty good idea, but as it seems that nobody is suscribing, they decided to offer a prize draw!!! great! What can you win? a Palm? your weight in fiber? a brand new router? no no no...The first price is an automobile sound system...the second price is....a rifle!! a winchester rifle?!?!? What the hell does that mean? I mean this is Lightreading, this is not "Car Tuning and Big Breast"...
I am sorry but I think this is cheap. If this is a joke, please let me know.

zipple 12/4/2012 | 11:02:53 PM
re: Hyperchip Lands $43M in Financing skeptic
If I may ask, what is your day job? Although some people here have different viewpoints, you have well articulated the VC viewpoint.
Myself, I am on the fence. From the viewpoint of a startup, having more high quality VC investors with skin in the game is a good idea. They all pull hard to make their investment pay off. (Yes, the company gives up some control. But that's why only high quality investors are desired.)
If the company brings an investor who is lacking in domain knowledge or business expertise (ie, govt), they might still bring to the table extraordinary purchasing power. (Yes it's dishonest and interferes with the free market.)

j-tripper 12/4/2012 | 11:02:50 PM
re: Hyperchip Lands $43M in Financing I saw the SuperComm box. It was a 160/320 gig box w/ OC192 cards present. There was a mix of ASICs and FPGAs.

Much of what 'sceptic' says about VCs WAS true. Today the winners are the survivors, not the startups with the best perceived VCs.

Some VCs are limited in the amount they can invest in any given company.

Anti-dilution clauses are pretty common, but usually pertain to the valuation going down.

IpOptical wasn't that big, so the amount of SW work being done there can't be that big.

What and how many companies in CN are involved in core routing? Probably not that many (Nortel probably did most of this in the US, remember Bay), the US move could be in response to this?

This type of SW scaling requires super computer experience not really routing experience. Again, the US move could be to suppliment the CN SW routing staff?

My two cents.

Belzebutt 12/4/2012 | 11:02:48 PM
re: Hyperchip Lands $43M in Financing What and how many companies in CN are involved in core routing? Probably not that many (Nortel probably did most of this in the US, remember Bay), the US move could be in response to this?

OPC software development was in Ottawa.

I think HyperChip is basically a "unique" company in Canada right now and especially Quebec, so that's probably one of the reasons they got the funding. And remember it's a more "socialist" country so government tends to sponsor industry a bit more. If you think that's bad, try Finland or France!

Re: the 65,000 OC-192 interfaces, is there any box out there in service today that has 65,000 of anything??? I mean, even a telephone switch in a captive office rarely has that many phone lines going into it. It would be a total nightmare to manage. Have you seen pictures of the HDX with a full line card load? That's just a fraction of what this Petabit router can allegedly scale to. And remember, OC-768 is still not here, it will only cut the number by a factor of 4, and it will take some years before it's economically feasible.

Why even bother quoting those peta-figures? They have to have a much better story than that to actuall sell the box, like "our software is extremely stable" or "our management is much more simple" or "our BGP can handle so many routes in real-life usage".
xxxx1455 12/4/2012 | 11:02:46 PM
re: Hyperchip Lands $43M in Financing After looking at their web site and some stuff posted here it looks like they just made a 16/32 OC192 Port GǣstackableGǥ router. You stack 2000 of them together and you get your 65,000 port system. I see it being a little better than stacking Cisco GSRs together, assuming that they have dedicated interconnect bandwidth and a distributed switch fabric. I hardly see it as a petabit router, anymore than I see the Cisco GSR as a terabit router. They will be hard pressed to have more than about 500Gbit/s worth of interconnect bandwidth which means that they are a multihop, blocking router, requiring multiple management sessions, before they even get to 1 terabit. Multhop means more jitter and latency for the IP phone guys. Blocking means reduced performance and less stability on the network (not good for the core). Mutiple management sessions means more cost and complexity to maintain. In other words, all the same old problems, no new solutions.

Not to say that itGs a bad product, I just donGt see it living up to its hype. Looks a lot like a Caspian play, but without requiring the proprietary protocols that the market wont adopt no matter how special you think you are. If they do get it out the door they might make some trouble for Cisco and Juniper, but only if the software is at least as good, and they can prove that these boxes truly stack together nicely without introducing any instability into the network. If they canGt, then they have nothing except maybe some cool switch chips to sell.

skeptic 12/4/2012 | 11:02:42 PM
re: Hyperchip Lands $43M in Financing Re: the 65,000 OC-192 interfaces, is there any box out there in service today that has 65,000 of anything???

They only quote that (silly) number because there
is a war going on between startups where each
tries to one-up the other in scale numbers to
the point where we get hyperchip's 65k number
and Caspian's "infinite scalability" claim.

What really matters though isn't always the
number, but their solution for getting there.
Caspian's solution was DMPLS and what appears
to be a mesh of their systems. Pluris is pushing
a multi-hop distributed switch.

I really don't think much of systems like that
because a big cluster-based router isn't
pratically speaking much better than room full
of meshed routers. The (bad) economics of the
mesh still apply. About the only thing you save
is some operations cost.

There is a need for bigger routers, but trying
to "fake" scale by building meshes or distributed
systems seems (IMO) to be a dead end.
stomper 12/4/2012 | 11:02:33 PM
re: Hyperchip Lands $43M in Financing > a big cluster-based router isn't
> pratically speaking much better than room full
> of meshed routers. The (bad) economics of the
> mesh still apply. About the only thing you save
> is some operations cost

Usually your posts are very well informed.
I strongly disagree with you on this matter.
To list the reasons why I would need a book.
Here is a summary:
A meshed set of smaller routers must use
line interfaces to create the mesh. This causes
several problems. The cross sectional bandwidth
of your mesh is uneven unless you use half the
ports on each router to connect them - quite
a waste. If you dont, your router is suseptible
to traffic patterns. Furthermore, depending on
the size and design of the mesh, you will need routers utilizing all of their line interfaces
just to connect other routers in the mesh -
a bigger waste. Finally, this architecture is not incrementally scalable (what if you want just
one more line?), fault tolerant (one line
going down can take out a lot of others),
or robustly managable (try configuring all
the routers in a big mesh).
Smaller logical routers that were designed to mesh together using non-line interfaces still have a
set of issues. The way most people are tackling
this problem is to use a fiber based crossbar.
This is not incrementally scalable (you need to
buy the whole crossbar), and it is not fault tolerant unless you use multiple fiber connections from each router to each other, an expensive proposition. Add to this the fact that fiber is
currently ~20x premium over electrical connection
on a bandwidth per pin basis - and you have a tough sell. The only advantage here is that you
can physically space out the smaller routers
using the VSR optics (might be good in smaller POPs).

So, IMO, a fully distributed singluar logical large router is far better in every category of merit than a mesh. Whether any of the implementations built so far meet this ideal, and whether these large terabit systems are needed yet, the market will eventually tell us.
krsr 12/4/2012 | 11:02:32 PM
re: Hyperchip Lands $43M in Financing skeptic --
>I really don't think much of systems like that
>because a big cluster-based router isn't
>pratically speaking much better than room full
>of meshed routers. The (bad) economics of the
>mesh still apply. About the only thing you save
>is some operations cost.

Doesn't it depend on how expensive it is to trunk the routers together? For chassis-chassis connections you don't need sonet. Using inexpensive interconnect instead of 192's that cost around $100k to tie the boxes together could make it worthwhile.

Then again, 10GE might solve that problem, and it depends on whether the interconnect is really cheaper.

skeptic 12/4/2012 | 11:02:27 PM
re: Hyperchip Lands $43M in Financing Stomper,

I agree fully with most of what you said. I think
what I said either came out wrong or wasn't

The point I was trying to make is that without
a whole lot of thought put into the product,
some of the cluster-based solutions degenerate
into systems that look a whole lot like a mesh
of routers. And, I suspect, will have similar
problems (many of which you pointed out).

I'm thinking specifically of people like Pluris
and caspian (based on what I know about them
which isn't complete by any means). Caspian
claims (I think) to solve the problem via DMPLS
and dynamic load-balancing inside the platform.
And I guess they will have their chance to prove
if it works or not soon.

The "fiber crossbar" style of router really
depends on what the characteristics of the
central switch are. I know a couple years ago
that people were talking about using things
(MEMS) which had characteristics ( setup
latency) that IMO made it impractical. But there
is probably better stuff out there today. I
also didn't think much of that project cisco
canned along these lines.

<<   <   Page 8 / 8
Sign In