Submitted by “USB 3.0 might be one of the big stories here at IDF, but Intel just showed off a glimpse of the future: Light Peak, an optical interconnect for mobile devices that can run as fast as 10Gbps. That’s fast enough to do everything from storage to displays to networking, and it can maintain those speeds over 100-meter runs, which is pretty astounding. Intel says the idea is to drastically reduce the number of connectors on mobile devices, which should allow them to get even smaller – but the demo was on a huge Frankenrig, so don’t expect to see Light Peak devices shipping any time soon.”
I know the article says
but it appears to pitch Light Peak as a peripheral interconnect.
I’m also aware that USB does have a networking spec (or at least, people have implemented networking over it). However it works poorly and is, basically, rubbish. It’s also rare.
So – why on earth aren’t we beginning to see a basic convergence of networking and peripheral interconnects? Why cant my mouse or keyboard talk over ethernet to the PC? Networking technology *exceeds* peripheral interconnect speeds (10Gbit Ethernet…), and it’s designed to be more robust, and work over longer distances (with speed fallbacks etc), and (transparently) different physical media. Admittedly power distribution on some ethernet specs is difficult but not a show-stopper (it’s something that both the peripheral world and the networking world would benefit from solving). One bus to rule them all!
Simple, Ethernet requires fiber or CAT5/5E/6 cables which are usually far thicker than your average keyboard/mouse cable, plus, the price of the MAC/PHY will most likely increase the price of low-end peripherals that don’t really require 10Gbps bandwidth. (Mouse/keyboard/webcam/printer).
… Though Intel might be using Ethernet as their carrier protocol. (The 10GbE figure looks suspicious)
Having said all that, I do agree that Ethernet (as a carrier protocol) can be used to replace a -lot- of proprietary protocols (USB anyone?) – heck, it can even be used to talk to high bandwidth expansion cards. (Yes PCI-E/MSI-X, I’m looking at you.)
As for Infiniband, game over people.
In less than two years, the price of a -fiber- SR 10GbE NICs went down from >5K$/port to less than 800$/port.
Unless something changes, you’ll be able to buy a UTP-capable 10GbE NIC for 100$ and pay 1$ for each 3ft of CAT7 cable in 3-4 years…
Sure, Infiniband 12x will most likely beat 100GbE to the punch, but you simply cannot beat the economy of scale.
– Gilboa
Edited 2009-09-24 23:13 UTC
Exactly – use ethernet as the base protocol (with the usual layering on top) and you can run it over whatever physical medium you like – at most you then only need a media converter (as opposed to needing something smarter like trying to run say Parallel over USB). Existing media standards could of course be reused for certain applications.
How is this different than infiniband?
Infiniband is a cheap, fast, flexible interconnect.
http://en.wikipedia.org/wiki/InfiniBand
From TFA:
Fewer, smaller connectors.
Longer, thinner cables.
Higher bandwidth.
Multiple I/O protocols on single cable.
Hmmm….
Infiniband and lightpeak are two completely different and unrelated technologies.
Infiniband is a networking technology for inter-CPU communication (usually used in clusters and supercomputers where the cost and trade offs makes sense). It is cost-effective compared to competing solutions, but a short infiniband cable will cost you $100 on sale and a hub will set you back $5,000-10,000. Definitely not cheap for “consumers”. Nor is it clear that Infiniband’s lean protocol specifies an easy way for devices to identify themselves or to pass data without encapsulating it in network packets (i.e., without requiring expensive chips in your devices/mouse/keyboard/etc.). Also, it’s not even clear that the technology could be easily licensed by chip makers such as Intel at a reasonable cost. It’s also a fairly complex connector as compared to USB.
You can’t just take a technology that’s meant for one thing (streamlined high-speed communication between “intelligent” computer nodes using a network in which expensive routers do the packet switching) and assume that it will magically work well for a diametrically opposite scenario (flexibly connecting many heterogeneous daisy-chained “dumb” devices where low-cost of the connector matters). Or else we would still be using telephone cables for everything…
Infiniband already has an optical physical layer. Also the technology was originally developed by Intel as a peripheral device interconnect but was overtaken by PCI as the preferred internal peripheral bus. Since intel already owns the technology it would not suprise me to find some of the IP elements of IB inside Lightpeak. As to connector and hub costs – the demo rig did not appear to be inexpensive and in the articles they are citing consumer deployments as being the road to low cost.
I wonder if this can easily be used for storage and clustering of several computers as wel ?
heres to hoping this fails miserably. why do we need this? optical is for long distance information sending. you can send just as much data on a copper wire when it’s that short of a distance anyways. why would you need a real long, and very fragile, cord between your computer and your mobile device when you could just use wireless? people want to use wireless. they don’t want to use a fragile cord. and if they need higher speed transfer that what wireless offers they can just use a metal cable to directly connect to send it at higher speeds