Hacker News Reader

Comment by Calwestjobs | Hacker News Reader

another insane thing is having 40 Gbps USB ports used as a networking cable between two PCs... Why there is still no project which can do this by simply installing trivial driver (in windows) and 4 line patch to enable this in linux? just charge 2 dollars per year and youll be milionaire (until LTT / MS catches up and steals it from you )

It is even possible to have linux machine act as a display port sink to be used as a capture card, for streamers, youtubers,... with 0 dollar investment, 0 hardware...

View on HN

> Why there is still no project which can do this by simply installing trivial driver (in windows)

The driver has been built in to Windows for years, as it's the same tech as Thunderbolt networking. Just plug two Thunderbolt or USB4

https://learn.microsoft.com/en-us/windows-hardware/design/co...

> and 4 line patch to enable this in linux?

Supported since kernel 4.15 in early 2018.

https://www.kernel.org/doc/html/v4.15/admin-guide/thunderbol...

> It is even possible to have linux machine act as a display port sink to be used as a capture card, for streamers, youtubers,... with 0 dollar investment, 0 hardware...

This is not true. When a USB-C connection is being used in DisplayPort Alternate Mode two or four of the high-speed pairs in the cable are literally switched from being connected to the USB controller to being connected to the DisplayPort controller. Both the source and sink devices have to support the same alternate modes to operate, so you would still need to have actual DisplayPort sink hardware to capture a DP-on-USBC signal. DisplayPort capture hardware is rare on its own compared to HDMI capture, I don't think there's a single one out there that takes a USB-C DP Alt Mode input because there's no real reason to ever do that.

win & linux - great ! Thank you. unfortunately not a lot of AMD systems with USB4 hw :/

display port sink - ah ok i thought it is some sort of tunneling. but it is just mux/switch + retimer. so only one way.

And for what it's worth, Macs have had this ability for basically as long as they've had Thunderbolt. The same can also be done over Firewire on all major platforms, but these days 400/800mbps just isn't what it was back in the day.

edit: Also meant to mention that while Thunderbolt has always been rare on AMD due to its requirement of an Intel chip, USB4 is somewhat common nowadays as it's officially supported on the Zen 3+ and newer CPUs

yeah i had mac and had to clone disk over tb one time.

but mac recovery is magical, wifi password is written in NVRAM and if you erased/replaced whole disk, "recovery agent" started and downloaded macos image over wifi and installed it ! it was on intel mac, im not sure if it still works same way in arm. it took 2 hours XD

apples firewire and tb had problems with security, it was possible to do DMA reads from your ram over those. from external device.

The DMA security issue was an Intel problem more than an Apple problem. It was Apple's popularization of Thunderbolt that got Intel to finally stop disabling IOMMU support for product segmentation reasons.

security issue was Apple thing, they say security, privacy but have tb, firewire with posibility of "TSA" agents cloning your drive in 3 minutes. has nothing to do with any product segmentation. also they HAD icloud backups not encrypted, but had big public fight with FBI about unlocking ones iphone... good old days.

> security issue was Apple thing

Not true; basically everyone shipping laptops with Firewire ports in that era was at risk of DMA attacks. Apple was merely the most notable vendor trying to offer something better than USB 2.0.

> has nothing to do with any product segmentation

Go look at which Intel Sandy Bridge or Ivy Bridge processors had VT-d (IOMMU) capability enabled vs disabled. The product segmentation strategy is on plain display. They made overclocking mutually exclusive with IOMMU for several generations of processor. There's no technological basis for that, just artificial product segmentation.

Please put at least a little bit of effort into fact-checking yourself before continuing ranting. It's not that hard, and you'll be much more convincing if you don't exaggerate your complaints to the point of being obviously wrong.

There have been many such attempts, but none ever really took off.

I think the core problem is that "transferring bulk data directly between two computers" is not a problem that many people actually have. At least, not in a way that isn't solved by other means like an intermediate USB drive or simply waiting for your conventional network to do it.

In reality, what you propose can be done with zero software: simply put two USB network adapters back to back inside your cable. That's traditionally how this has been done. If you want a raw USB connection, one device must support USB device mode, which is extremely rare on PCs. I don't know if you can fix this in userland, or even the kernel. My bet is that this is a hardware/firmware level feature, though I don't know for sure.

i was thinking about something like using NAS to USB dock my notebook / tablet.

power delivery (even thru some kind of splitter)

+ 20/40gbps network for syncing AND internet (wifi is constantly down rating connection which does annoying delays, latencies, i live in high rise...)

It is somewhat possible but there's rumors that Intel intentionally nerfs the speed so as not to compete with their datacenter products. It could also just be some other resource limit in the hardware, you usually aren't saturating a thunderbolt connection with just one type of data stream. This guy built a small ring topology network using mini PCs and thunderbolt 4.

https://fangpenlin.com/posts/2024/01/14/high-speed-usb4-mesh...

The distance isn't great for USB/Thunderbolt 4 cables so it would really only make sense for cluster computing and since no switches exist for this kind of thing, you're limited in network topology. Generally, the kind of PCs that you'd need a fat pipe connected to already have the slots and lanes available for 40G or 100G cards. It's a cool experiment for the homelab but it's clear the industry isn't interested.

im not sure about intel limiting. a lot of times people forget to calculate encoding and/or packet headers into raw speeds. or something like that.

or in that article it is possible usb4 controller is sharing bandwidth with some other device or it is only 20gbps PER controller and he has connected 2 cables to same computer in middle. so 10 IN + 10 OUT is 20gbps ? maybe if he tries to measure speeds for only neighboring machines it would be 20gbps ? but his speed is 11.8 so does not make sense either. i do not know.

i was thinking for home use, when i come home with notebook i can quickly sync it to nas. both have NVME drives so 3-4 GBps should be possible. especially video. also not needing wifi, because im connected thru same cable im charging is awesome.

also recovery/ssd cloning is nicer on quick connection.

virtual machines can stay on nas too with those speeds. no need to copy.

>Why there is still no project which can do this

>just charge 2 dollars per year and youll be milionaire

I don't know, why don't you do it? Sounds like easy money.

there is not enough people to do stuff, in any profession.

USB isn't very reliable in my experience. It could work if you add your own error correction on top.

You can deal with that on another OSI Models layer. Just encapsulate/transfer something which already has FEC. not sure if wireguard has fec

You want your FEC at the lowest level possible in the OSI model so errors are detected early in the stack and you resend as few packets as possible.

LTT?

youtuber which spreads marketing misiinformation and laugh into your face on live stream while doing it.

Reply on news.ycombinator.com