Embedded world...

[juanplacco]

BTW, is there someone using QNX outside the embedded world?

JM

[ianc]

Sure. We use QNX (4.25) as a server platform for our Security/Access Control Management Software.

[juanplacco]

Thanks for the reply.

Someone else??... Someone using QNX 6.* ?

Regards,
JM

[mario]

Most people I know do not use it in embedded system

ADDED: Maybe you should describe what you mean by embedded ?

[maschoen]

I use QNX 6 for development (Duh!) but also as an Internet server.

[juanplacco]

Embedded != "Some mother board into a big case + some powerfull uP with a big fan cooler + lots of PCI slots with many plugged cards doing something + big hard drive (maybe a raid) + + + KVM +++ " and finally... many of this "packages" (Industrial PCs (maybe 20)), working together into a single (qnet?) network.

well... ...

[mario]

we are down from 20 (qnx4) to 3 pc ( qnx6 )

[Tim]

we are down from 20 (qnx4) to 3 pc ( qnx6 )

Ah, the power of Moore's Law + Multi-Cores.

We need only 2 QNX PC's (one PC104 board on the vehicle and one industrial rankmount in the control center) for our real time work (the GUI's are done on Windows). Furthermore we don't have any PCI plug in cards. All our I/O is done by external boards (custom designed + 3rd party) communicating over serial or ethernet (this saves the trouble of writing drivers).

Tim

[maschoen]

Has anyone noticed that Moore's Law technically (if not effectively) has slowed down. Otherwise we'd have 8 or 16Ghz PIV's by now.

[mario]

Well I think it still applies because although you don't have higher frequency, you have multiple core. AMD came out with a 12 Cores processor running at a maximum of 2.3GHz so that's 27.6 GHz

[maschoen]

Well that's what I meant by effectively. The key of course is that the cost per core must continue to come down if the speed does not go up.

[Tim]

Also it's worth mentioning that the theoretical 27.6 Ghz is only possible if you have a threaded program that can take advantage of those cores. If you don't (say a PC game or formatting a Word document) then CPU speed hasn't improved much in the last half dozen years.

It's also obvious that Moore's law can't continue indefinitely because it's a geometric rate of increase that would have processors with virtually infinite speed/cores by the start of the next century.

Tim

[maschoen]

quote="Tim"]Also it's worth mentioning that the theoretical 27.6 Ghz is only possible if you have a threaded program that can take advantage of those cores. If you don't (say a PC game or formatting a Word document) then CPU speed hasn't improved much in the last half dozen years.
[/quote]

Yes but note that in the case of formatting a Word document, once it takes less than a .1 second to format a 10,000 page document, it doesn't matter if you can speed it up or not. If you have 10,000 of these documents to format, well that's where multi-core comes in handy. The only place where you run short is a program that must run linearly and for which there is always a gain for running it faster. I can't think of any right now. There are probably some mathematical problems that work this way.

It's also obvious that Moore's law can't continue indefinitely because it's a geometric rate of increase that would have processors with virtually infinite speed/cores by the start of the next century.

Tim

If we allow for increasing numbers of multi-cores in exchange for increasing processor speed, I think the end of Moore's law will be market drive. There will be multi-core devices that are cheap and do pretty much whatever anyone would want to do already so why make anything better. Imagine the volume of an iPod filled with multi-cores.

[juanplacco]

The problems are temperature and the size of an atom. We need a hardware revolution not based in transistors...

[Tim]

The only place where you run short is a program that must run linearly and for which there is always a gain for running it faster. I can't think of any right now. There are probably some mathematical problems that work this way.

Doesn't sorting fall into this linear category? The best sort algorithm is Quicksort which works by dividing the number of items into ever smaller groups that themselves get sorted. Now to take advantage of multicores (ie each sub group gets assigned to another core to allow parallel sorting of these sub groups) requires a software rewrite in addition to the hardware core increase. It's not clear that rewrite is happening or is going to be happening except in special software (like 3D Studio Max) designed to benefit from multi-core. In a regular program it's much faster to sort with a single core CPU with faster speed than a multi-core CPU with theoretical faster speeds.

If we allow for increasing numbers of multi-cores in exchange for increasing processor speed, I think the end of Moore's law will be market drive. There will be multi-core devices that are cheap and do pretty much whatever anyone would want to do already so why make anything better. Imagine the volume of an iPod filled with multi-cores.

I don't know about you but I'd be much happer if my iPhone had multi-cores in order to run more than one app at a time

We need a hardware revolution not based in transistors...

how about a software revolution based on less piggy code

Tim