4 ms·
I've got a couple of the Chinese made "Remote USB" servers which are interesting but not as compatible as I would like. What I find really amusing is that back
by ChuckMcM 11y ago
I've got a couple of the Chinese made "Remote USB" servers which are interesting but not as compatible as I would like.
What I find really amusing is that back when USB was proposed I argued that rather than create an entirely new infrastructure why not just use Ethernet packets. There are additional protocol types available in the bit encoding and all sorts of infrastructure that already supported layer 2 switching. Plus the "phy" (10/100 at the time) was dirt cheap and already available. It never went anywhere, and at the time I figured that nobody wants to support a standard that they can't see how to make licensing money off of it. Somewhere I have VHDL code for a PC parallel port emulator that ran over my protocol. It fit in a small FPGA with the ethernet MAC so it wasn't like we were talking about a lot of gates. Just no consortium fees and no way to keep people out of the ecosystem you didn't like.
Doing USB over IP is an interesting way of re-creating that work (although mine was multi-master so technically a bit better :-)). Of course these days USB 3.0 is faster than the ethernet port on most machines so it was no doubt doomed from the start. But its fun to do. The SATA over Ethernet stuff that CoRAID did was like that, very nice work.
- jjoonathan 11y agoIt's been funny to watch the various phy standards converge in feature-space (alas, not interop-space) over time. PCI got a packet architecture with addresses and routing and credit-based flow control, USB got a packet architecture with addresses and routing and credit-based flow control, SATA said "screw it, we'll just become PCI-e, because they have packets and routing and flow control," and thunderbolt basically is PCI-e... Unfortunately I think Ethernet might be too fragmented to ever take the convergence to its logical conclusion. But who knows. If someone high on the food chain at Intel wakes up on the other side of the bed one day it could happen.
- simcop2387 11y agoEthernet's also got the fun problem of the length of the runs being hugely different than pcie. What I expect is some kind of optical interface because of that. Optical thunderbolt might be the way for that to happen.
- jjoonathan 11y agoI thought 40G/100G ethernet was already optical (and multiplexed, much like PCIe)? EDIT: yep https://en.wikipedia.org/wiki/100_Gigabit_Ethernet https://en.wikipedia.org/wiki/100_Gigabit_Ethernet I'm pretty sure that the only obstacles against "ethernet as a truly universal bus" are either minor (someone's got to invest in making adapter interfaces) or of the "why bother" variety. Unfortunately the latter is a bit of a show stopper as far as I can tell :)
- simcop2387 11y agoHadn't thought about 40G/100G ethernet. It'd definitely be in that same area for doing that. I'd definitely agree that the obstacles at this point are either minor or a "why bother" kind of thing. With speeds continuing to go up, I'd expect it in a generator or two simply because the "why bother" will shift the other way, why bother to make a new standard when this one does it too and is nearly ubiquitous. My guess at least for peripherials is going to be in the optical thunderbolt area, more because of the way the device network matches what people are designing for anyway. I'd guess once there's two standards, a local (optical pcie) and remote (optical ethernet), things might begin converging more and more until it's less distinct between the two. I'd expect optical more because it's easier to get high bandwidth and low latency on interconnects than copper because of the lower crosstalk and everything. It's just a question of market forces pushing things to converge.
- TazeTSchnitzel 11y agoThunderbolt is PCI-e... plus DisplayPort, because we need two different serial protocols! (And it frequently carries USB and Ethernet. So many protocols.)
- TD-Linux 11y agoThe Ethernet phy is actually pretty expensive, especially compared to USB 1.0. The isolation transformers alone are $1. Of course you could skip the isolation like some backplanes do. The ID assignment for USB is pretty awful and so is the robustness - for larger products it seems Ethernet is on the rise again.
- spin 11y agoI feel the same way (that ethernet is a perfectly good way of moving data around), so I created this a while back as a side project: https://github.com/mfassler/eFirmata Basically, I wanted USB, but I wanted much longer cable-lengths than USB, so I used ethernet.