IDF San Francisco |In an attempt to break the USB bottleneck, Intel has kickstarted work on USB 3.0, aka 'Super Speed USB'. First devices aren't expected 'til 2009 though.
USB is one of those rare pieces of technology which actually lives up to its name. From PCs of all shapes and sizes, to peripheral devices, mobile phones and even an increasing number of living room-based consumer products, it's truly become a ‘universal' connector.
However, it's also becoming a bottleneck when it comes to moving large files from fast-moving drives - particularly capacious flash-based solid state drives which are moving to the fore in the mobile device space, including forthcoming digital cameras and camcorders.
While the first day of Intel's annual San Francisco gabfest appeared focussed on processors, a breakout session with surprisingly little fanfare announced the third generation of the Universal Serial Bus spec, to be developed by the USB Implementers Forum. Intel is not only one of just six companies with a seat at the USB-IF's director's table, but its "technology strategist" Jeff Ravencraft is Chairman and President of the club.
While not pegging a peak speed for the nascent spec, Ravencraft told apcmag.com that USB 3.0 would offer at least "over 10 times performance increase over USB 2.0" - which translates into at least 4800Mbit/s, or 600MB/s. That's the theory, anyway - as is the case with current USB specifications, and life in general, reality doesn't always match up.
Intel software architect Rahman Ismail, who shared the stage with Ravencraft, said that while "with USB 2.0 you can get 40MB/s on a good day if the stars are aligned", USB 3.0 would still have "a guaranteed throughput of 300MB/s... we have flash devices hitting 45-50MB/s, so we need to provide the headroom so that USB doesn't become the bottleneck".
Part of this will entail a new ‘mass storage protocol' which allows USB-based devices to act as hard drives. "Current USB mass storage averages around 32MB/s, and the existing mass storage driver will hit (its limit) at 150MB/s. We need a new mass storage protocol, and our goal is to hit around 400MB/s" said Ismail.
The spec's design would also "minimise software overhead and wasted bandwidth" which co-ordinates data shuttled over the bus, as well as remove the constant polling of devices to check if they have data to send to the host PC.
"In USB 2.0 you're always asking do you have data, do you have data, do you have data. But even with a mouse or keyboard, 90% of the time there's no data, you're sitting there thinking". USB 3.0 does away with polling, says Ismail - "when a device has data, the device will transmit the data. And if you're idle, we want (USB 3.0) to go to the lowest power state possible, not keep polling the device to see if it's in use".
This not only reduces the power drain of a USB device when connected to a notebook, it also cuts back on traffic and ensures devices that do have data get immediate attention. "When USB is checking if your mouse or keyboard has data, it's not checking if another device has data". The spec will also support native virtualisation of USB 3.0 devices, says Ismail. "This lets a virtual machine interact directly with a USB device, without any software intervention, such as storage devices with a partition for each VM."
As you'd expect, USB 3.0 will be backwards compatible with USB 2.0 and will use the same connectors. However, "the insides of the cable will be different, but the user won't notice" said Ravencraft. The revised cable "will have dedicated in and out lanes, so you can get data from a device while moving data to a device at the same time. You can have data streaming in both directions at the same time".
Following the branding of USB 2.0 as ‘Hi-Speed USB' - which we've always found confusing, considering that USB 1.1's snail-like 12Mb/s pace is dubbed ‘Full Speed' - USB 3.0 will carry the consumer brand of being ‘Super Speed' USB. But the devices themselves won't arrive Super Fast.
"We intend to have a completed spec in the first half of 2008" predicts Ravencraft. "When the spec gets to the .9 level people will begin to build products, so we'll see product development through 2008, with initial deployment in 2009 and broad deployment in 2010".
David Flynn is attending IDF San Francisco 2007 as a guest of Intel Australia.