I think Light Peak (LP) is useful and will probably get some market traction. It is the hand wavy claims that it going to eliminate all the other protocols that is just hype and smoke.
I do grasp it. Managing a protocol by simply transporting the signals from one box to another does
not remove , change , or supercede the protocol. A USB 2.0 connection routed over LP will be just as fast as USB 2.0 over a normal USB 2.0 cable. You can get complementary attributes to USB by using LP. For example you can extend the distance between the hub and the device ( presuming provide power to the client side if necessary). Likewise you can run USB 2.0 and a Display bitstream over the same wire ( and decode on the both sides back into the underlying protocols). So you get fewer cables.
As long as it the diagram goes from something like.
[usb controller ]------< usb cable >-------[ usb client ]
to
[usb controller]--[LP controller ]---<LP cable >----[LP controller]---[usb cleint]
you are not removing anything. You are only
adding components to the system. That will almost always make it more expensive.
Note that as long as the computer still has a usb controller, PCI-e controller, etc. in it then it is trivial to just run a physical connection out to a single protocol specific socket. if there 10 USB connectors, 5 SATA ones , etc. then if you just prune a few off to LP routing, you can still use the rest with extremely minimal cost out to the side of the box.
There is a couple of general approaches Intel could implement this multiprotocol support. One is internally add the legacy ones to the controller. The other is something that just decodes the signals which are routed somehow to the legacy controllers ( by pins and/or other means.)
Neither one of those removes the controllers from the system. All the current properties are still there. You just get complementary properties that LP enables (e.g., distance , fewer cables , room for future compatibility with updated standards with speed improvements if added. )
I don't really buy that first option will be widely taken. If that happened at all it would be way out in the far distance future. Likewise it is extremely doubtful it would happen across the board.
This is a variation on the classic "Embrace , Extended , Extinguish" methodology that folks like Microsoft and others have attempted to use.
The problem for LP is that the mixed use will be the normal mode. As long as the legacy protocols are supported in the core chipset they will always be around.
The other major problem is that a couple of the legacy protocols are fast enough. For devices with dramatically less than 3-5Gbps bandwidth requirements all that extra speed doesn't buy much of anything. Especially if it is only close distances to a singlular device you are trying to traverse (e.g., a small single chip USB Flash drive plugged into a socket on the computer. )
the claim that are going to subsume and eliminate has been trotted out many times before. For instance many years ago written about Firewire:
http://www.ausairpower.net/OSR-0201.html
SCSI is still here. It mutated into SAS and iSCSI but still around. I don't think Ethernet , USB , SATA , PCI-e etc. standards committees are just going to throw up their hands and say "we give up.... Sauron's one true ring, LP, will rull them all." Nor will having LP making you automagically forward compatible. If all you have internally is a PCI-e v2.0 controller and there is a PCI-e v4.0 device on the other end ... you are not going to get "v 4.0" in your "2.0" box just because they are connected by LP.
Ethernet can use the exact same optical transceivers that LP is using and already has 10-100Gbps standards. Likewise PCI-e and SATA can pick up optical foundation without picking up the LP protocol overhead. There is a huge pile of USB devices that just don't need the speed at all. 50Gbps LP .... phfffff printer stil runs at exactly same speed.
A decent portion of the LP spin is just hype to get folks to adopt yet another new standard protocol. If they came straight out and said "we just want everyone to adopt something new so you have to buy all new stuff". There would be a very large "forget you" and the protocols would continue to evolve.
So Intel spins it as .... "Oh we are going to get along". Then once have traction enough to get included onto a high enough percentage of boards then get folks to do native LP protocol devices. However, it is extremely dubious that will eliminate all the other sockets. It is far more likely to get LP to the point where it can get enough traction to survive, not subsume , the other protocols. In short "embrace and extend" will work well enough. The " Extinguish...." that isn't going to happen.
There are too many things that Ethernet (the world wide web ) , Infinibate (low latency transfers ) , SATA ( high enough bandwidt) , etc. that LP just don't have a better story for.
Basically for the exact same reason that just about everyone still has a RJ-11 jack in their house (even if have VOIP). 25-30 years about phone companies used stuff like microwave, cables, sat, etc. to move phone traffic. Over time lots of that go multiplexed on to fiber cable. That didn't remove the RJ-11 jacks from folks out. Adapters flips the bits back to the legacy traffic so that still plug into the same old jacks.
(yeah there are internet only VOIP phones, but they are small, small , small percentage of all the phones in use. It will be many years before that changes. )
