[prev in list] [next in list] [prev in thread] [next in thread] 

List:       linux-firewire
Subject:    [Linux1394] polishing a turd
From:       "Gary A. Hoffman" <hoffman () ricochet ! net>
Date:       1999-07-08 23:51:08
[Download RAW message or body]


Can USB be any more than a PC-centric interface?
Maury Wright, Technical Editor

Working on the home-automation and networking articles in this issue reminded
me of an opinion I
meant to express earlier this year when Intel unveiled USB (Universal Serial
Bus) 2.0 (www.usb.org) with
proposed extended data rates to 120 or 240 Mbps.

What's the connection? Well, the emergence of home data and control networks,
the convergence of
computer and consumer electronics, technologies such as Jini and Universal Plug
and Play (UPnP),
ubiquitous portable wireless devices, and numerous other factors point toward a
decentralized,
distributed-intelligence computing/communications model for our future in both
the home and the office.
Conversely, with USB 2.0, Intel is trying to extend a PC-centric technology
that, in my opinion, was
poorly conceived to start with. The bad news is that Intel might be just
persistent enough and wield just
enough influence to slow progress toward a more heterogeneous future in which
the communications link
is more important than the node itself.

I admit that a single serial bus that connects all PC peripherals is a good
idea if you view it strictly from
a PC perspective. I wish we had such an interface 10 years ago when 12-Mbps
rates would have been
truly useful. Intel, along with other PC-industry players varying in their
enthusiasm, still think that USB
can fill a universal role. But this scenario presents two major obstacles.
First, USB has limitations that
prevent it from being a universal serial interface for the PC, and even a
face-lift won't solve these
problems. And second, from a chronological perspective, the future simply can't
tolerate a PC-centric (or
any host-centric) interface or bus as the preferred interface for many new
products.

USB has already weathered a multitude of problems and surely would have
disappeared without Intel's
persistent backing. Many blame Microsoft for delaying OS support, but the
hardware spec also had
significant problems. Today, USB 1.1 is just overcoming compatibility problems
that stem from both
hardware and software issues. The compatibility problems will return and maybe
intensify as Intel tries to
support 1.5-, 12-, 120-, and perhaps 240-Mbps rates on the same cable as USB
2.0.

But assume for a second that compatibility happens. USB still has performance
limitations. The
12-Mbps rate is a clear limitation that largely prompted the USB 2.0
initiative. But data rate isn't the only
issue. The interface depends on the host to initiate every transaction, thereby
adding latency, especially
as you increase the number of devices connected to the bus. Tests by EDN
editors working with
industry experts indicate that USB sustains just over half the maximum rate.
USB proponents claim that
new USB ICs let the interface support rates closer to the maximum. But all
these discussions assume
one host and one target. The bandwidth available to any one device on a loaded
bus is a fraction of the
maximum.

Presumably, a move to 120- or 240-Mbps maximum rates will solve the problem.
But, realistically, the
latencies of a host-centric interface and tens of peripherals won't disappear.
USB 2.0 will be no more
suitable for data-storage devices in the future than USB is today. Moreover,
USB 2.0 still won't support
the peer-to-peer communications that high-end multimedia devices, such as
cameras, require. And it's
still unclear whether USB 2.0 will ever be considered isochronous and will be
capable of carrying live
video.

In fact, it's still unclear, from a technical-detail perspective, whether USB
2.0 will work at all because the
first version of the spec won't debut until late this year. USB 2.0 systems and
products are slated for the
second half of 2000. Originally, Intel claimed it could roll out USB 1.0 in
about 18 months. It took more
than four years, and problems still exist. By the time USB 2.0 is viable, even
higher data rates will be
necessary and host-centric interfaces won't be appropriate for scanners,
printers, cameras, external disk
drives, digital-versatile-disk drives, tape drives, and other devices.

Consider two of the most compelling software initiatives for the convergence
era, Jini and UPnP. Both
initiatives will allow devices such as scanners, cameras, and printers to carry
their own device drivers.
Presumably, a scanner will connect directly with and send output to a printer
without host intervention.
Such technology will require a peer-to-peer interface that operates at rates
far beyond 240 Mbps.
Fortunately, a good choice already exists-IEEE 1394. This standard is up and
running today at 400
Mbps and will hit 800 or 1600 Mbps by the time USB 2.0 devices ship. Moreover,
1394's originators
designed the standard from day one for peer-to-peer and isochronous operation.
In my opinion, IEEE
1394 would already be in most of today's PCs if Intel hadn't gone all out to
prevent such proliferation. In
the heterogeneous future, Intel will ignore the standard at its own risk.

As for USB, I have one great hope: that my son will be able to keep all of his
PC devices-gamepad,
steering wheel, joystick, digitizer, and others-simultaneously connected rather
than expect me to
switch plugs at his whim. But like one of my good buddies from Georgia often
reminds me, you can
polish a turd all you want, and it's still a turd. I just can't see USB being
any more than a PC-centric
interface that replaces mice, keyboards, game devices, and, perhaps, audio and
modem ports.

gary a hoffman

hoffman@ricochet.net

_______________________________________________
Linux1394 maillist  -  Linux1394@eclipt.uni-klu.ac.at
http://eclipt.uni-klu.ac.at/mailman/listinfo/linux1394

[prev in list] [next in list] [prev in thread] [next in thread]