Slashdot Mirror
USB 2.0 For Linux
SilentTone writes: "PCWorld is reporting that USB 2.0 or high speed USB will be hitting Linux first half 2002. Intel is already providing space on its Pentium 4 motherboard for the USB 2.0 controller. With a transfer rate of 480Mbps (more than firewire's 400Mbps) it seems promising." Update: 09/04 23:02 PM GMT by H : So, somewhere between my preview and going live, I seem to have "lost" the link - if you find it, please post below. I'm looking - in the meantime, this is a good Linux and USB tutorial, and Blue Cat Linux is supporting USB 2.0. HA! Found it - story updated.
At any rate, Linux support for these next-generation devices is still important; better for it to come sooner (before it's popular) than later (at which point people wonder why Linux is lagging behind).
David E. Weekly
Code / Think / Teach / Learn
h4x0r for
After a few minutes of searching I think I finally figured out exactly where PCWorld has "reported" that Linux will have USB 2.0 drivers in first half of 2002. It is located in this story .
Here is the information they give:
"But don't count USB 2.0 out. Microsoft has announced that it will offer downloadable USB 2.0 drivers for Windows 2000 and for the upcoming Windows XP operating system. Linux support for USB 2.0 should come in the first half of 2002.
Silicon behemoth Intel currently provides space for a USB 2.0 controller chip on its Pentium 4 motherboards, and Gateway has announced that it will put the chips in some PCs beginning this fall. Intel and Acer Labs plan to put USB 2.0 into at least some chip sets by mid-2002; Via Technologies, on the other hand, will add IEEE 1394 support to its chip sets before turning to USB 2.0. AMD says it will support USB 2.0, but not how or when."
Sounds like speculation to me on the Linux drivers. Do any Linux USB devs have any actual info about this?
Q.
I think maybe he means http://www.pcworld.com/news/article/0,aid,60124,00 .asp.
The theoretical transfer rate of USB is misleading. Overall, USB remains an inferior technology to USB for applications requiring high bandwidth with deterministic, isochronous transmission. This article provides a good explanation of some of the issues involved. In one of the projects I have been leading, we have been involved in developing the Linux IP over 1394 drivers, and have obtained over 150 Mb/s point-to-point bandwidth using IEEE 1394 asynchronous mode, with room for left for further optimization . The increased function call overhead of USB makes even this modest performance level unlikely.
We are saddled with this inferior technological solution due to the recalcitrance and greed of Intel, who, as usual, are elevating their hidden agenda borne of backroom deals and "strategic partnerships" above the interests of their customers.
Agreed...
People keep saying that USB 2.0 is faster then 1394a and it is, slightly. 1394b has a number of other benifits that USB doesn't have.
1. 1394 is non-computer centric. There can be any number of computers in a chain from zero to 63.
2. 1394 provides an isosyncronous mode of transmision. This is required for streaming video.
3. 1394 has better methods of bandwidth enforcement then USB
Beyond thoes benifits, 1394b supports speed up to 3.2Gbits/sec at up to 100 meters over fiber.
Another sign from the industry is that theLucent spinoff, Agere Systems, has scrapped plans to produce a USB 2.0 chipset and in order to speed development of it's 1394b chipset http://www.lucent.com/press/0701/010716.mea.html
USB 2.0 looks to be too little, too late.
I bought in January of '97 a Tyan Tomcat motherboard with a P75 chip, it had a USB port. The spec changed and USB was busted, so they relabeled the parts (and replaced the motherboards if you complained enought).
The problem wasn't motherboards. By mid-97 all machines had the ports. Every machine my high school picked up that year (first half of 97) were P133s-P166s and had USB ports. These were Dell computers with vanilla mobos.
The problem was originally software support, MS didn't support USB until Win98, the Win95 support was busted. Additionally, the market for mice and keyboards died around then. The computers shipped with them AND the market was only supporting $20 replacements, not the highend ones that were popular through 96 and early 97.
Without software support, there was no interest in the hardware. People were pushing parallel port solutions instead. The parallel port scanners, zip drives, etc., dominated the low-end, and SCSI still ruled the high end.
Apple made USB a reality. They used it to replace ADP when they needed something to replace the external SCSI-1 port they used for expansion forever. With their move to IDE hard drives, the SCSI port was rediculous.
Anyone selling addons for Apple built them as USB devices, including mice and keyboards. As the standard was the same, there was no reason to not write Win98 drivers and open up the PC market.
Apple's ability to make something a standard on a segment of the industry is powerful. While Dell and Compaq (soon to be HP) ship lots of machines, nobody is interested in a Dell-only or Compaq-only option on the consumer level. The PC world is commodity only now, so only MS/Intel can add things to the standard. There is no room for vendors to improve the experience, since we scream and yell that it is propriatary.
USB 2.0 is a bad hack. If you don't use a USB 2.0 hub, then any USB 1 device (which keyboards, mice, scanners, etc., should always remain) drops the whole thing to USB 1. In addition, the bus is split up, so the 480 MB theoretical is a real joke. The bus uses time slices, not bandwidth slices. So when the keyboard and mouse grab their fractions of a second, they take bandwidth that could go to the video camera.
Furthermore, Firewire 2.0 brings Firewire up to 800 MB, and its reality is much closer to the theory.
OTOH, I agree that it is good for Linux to support it. As Linux distributions/kernels in the wild don't get upgraded as often, having the support now means that in 2 years, everyone will have it. Better to have the software beat the hardware.
Adding support in Windows is more user-painless (insert CD, press setup, watch this application you got from a no-name vendor to save $3 overrights basic operating system files...) then Linux, so it is good to see Linux beat the curve.
Alex
There is an EHCI driver and USB 2.0 core patches available right now for Linux.
The current plan is to merge them into the 2.5 kernel, and perhaps backport into the 2.4 kernel once it is deemed stable.
The problem holding back USB 2.0 under Linux is device availability. We've had a couple of vendors donate some USB 2.0 Host Controllers but only 1 device. There are a couple of devices available for purchase now and they work with the aforementioned patches.
The story on pcworld.com is speculation. We have USB 2.0 support, but it's not finished and it will only be finished when we have devices to test against.
And not to put too fine a point on the excellent comment that "it needs a computer to function" -- USB puts *ALL* intelligence in the USB host.. i.e. the computer. If you're Intel, this sounds like a good thing. If you're an embedded developer with anything resembling a processor, it can actually rather suck.
You see, the USB bus basically requires that the host POLL the client devices, with all of the problems that entails. E.g. the USB device's driver polling frequency determines the latency by which the host can accept an update of client state, and since polling itself requires client processing bandwidth to handle, polling faster sucks more client bandwidth. Moreover, polling sucks cycles even if there is nothing to tell the host!
USB is great as a simple desktop bus facility since it does permit cheap implementation of things like keyboards, mice, etc. But for high-rate communication between embedded devices USB is just awful.
Or a single user system with more processes than just the benchmark program running.