Intel gets cold shoulder from Orange delegates at Code Camp
Not ready for message
"This place is full of Java programmers. They don't care about optimising. Well, they do, but there's nothing they can do about it." Thus "Dr Dave" Layman, jokingly consoling himself (and Intel) for the empty classroom for his lessons in how to make Intel processors go faster by using co-processors.
Intel's classroom empty? What about Microsoft's session? Hang on, where exactly is Microsoft? - not even here?
What on earth is going on?
It's day two of Code Camp, the outdoors tech-fest for Orange developers from all around the world. We're in a virtually empty room in Futuroscope, near Poitiers, South of Paris, and we're studying.
The previous session had been a deep-structure revelation of how 3G phone networks operate - right down and dirty in the protocol carrier and machinery layers, missing nothing except the basic radio technology. And that session was packed. Coders listened intently as John Faulkner, Principal Designer of UMTS architecture, described how the GPRS network looks the same as UMTS, and also where it differs.
For example, he revealed exactly why you'll get better performance for WAP users over GPRS than UMTS if they're opening a page. It turns out there's a big delay in setting up the connection in UMTS, and the UMTS "bearer setup time" takes as long to complete as it takes to send about 50K over a 2.5G phone link.
Layman (PCA ISV optimisation lab manager, Intel) gave a good session, in my non-expert view; well worth attending. But the audience of 250 developers at Orange's Code Camp was not interested in optimising for any one specific platform. They want to write one application, and deploy easily across the widest range of phones. And Layman's quite right: today, that means writing in Java, or at the very least, a general high level language.
What Layman was showing was how to write code to invoke two of the co-processing units Intel has built onto the basic ARM core. One is iMPT (Intel Multimedia Processing Technology) and other is our old friend MMX, reborn as Wireless MMX.
His trick was to take a basic signal processing operation, and show how it would run on the naked ARM chip, then introduce calls to these co-processors, and running benchmarks to show how this speeded the process up. And then he showed how you had to write the code, and then he showed a few FCCs (frequently coded crimes) and how to avoid getting caught.
Standing room only
But the session which attracted the "standing room only" signs was Faulkner's. What made Faulkner's 3G phone presentation so valuable, was the way he was able to candidly reveal where the system falls short, and how to work your way around it while system fixes - already sorted in the standards lab - fight their way through the system into equipment that will be in the field in 18 months' time, or thereabouts.
For example, the designers of the 3G system originally tried to make it radio transparent - and succeeded. "Maybe they shot themselves in the foot," commented Faulkner - that is, maybe they shouldn't have. Transparent means if you're sending data across the network to the device, you don't have to ask whether the device is connected over 2.5 or 3G. Which is great, but unfortunately, you often need to know that.
Video, for example, won't stream to a 2.5 G phone the way you can do it on 3G; the speed isn't the only problem, because there is latency, too. Is there a workaround? Yes! - Faulkner showed how Orange tests a system on its own network to see which Access Point the message is coming through, and - from that, what sort of radio it is using.
Which is great when you're on Orange in the UK; but when you go roaming, all that Access Point Name (APN) information is lost. And he went through other issues like that, and the audience was very, very focused, because they're looking to deploy applications on 3G networks over the next year, and will need to know what sort of things go wrong.
All these problems will be sorted out - in a year or more. The issues, Faulkner assured us, have been taken on board, and the standards bodies have written new requirements - but those requirements aren't built into the equipment in the field today. And nobody is going to upgrade them for the pitifully few users there are for 3G today, unless there's an instant return on investment, are they? No, they aren't...
Intel will have to wait a while before anybody really cares about the story Dr Dave has to tell. One day, no doubt, people will want much faster phone performance. Today, they're far more concerned with ease of installation - and having to tell the network what make of phone you have, and what software release, and what chipset version, is the sort of carry-on that discourages ordinary phone users, drives tech support offices up the wall, and in general, really doesn't look like a market.
And of course, Microsoft, which rides on Intel's coat-tails in portable computing, seems even further from anybody's mind.
But that's another story, and an odd one.
Sponsored: Hyper-scale data management