Post a Comment
This should speed up installation quite a lot. Lack of DMA support was slowing down the unpacking of packages a lot.
It looks like the next release will be a major upgrade compared to current release
http://www.skyos.org/?q=node/447
Edited 2006-10-12 16:00
> Have you ever USED SkyOS? It ran plenty fast even with
> PIO.
I have, I paid for it and... well, it was pretty damn slow. Installation took over one hour here, and opening just about any programs used to take a very long time.
Also - albeit unrelated - the gui was pretty slow as well, and the mouse a bit jerky.
Expecting ANYTHING to be full featured in day 1 is just arrogant, development takes time... They even specificly mention that SkyOS is still BETA on their page.
Anyway, there was a problem and they fixed it. That's what we call improvement. It doesn't matter if they had it five years ago or not, it's what they have now that matters.
Hmmmm...How fast can YOU write a driver for any of SiS, Intel, Via or some other chipset that supports everything, including DMA? 1 day? Somehow I *very* much doubt that!
Implementing DMA isn't that easy as you think it is. The basic idea behind it can be explained easily, but it is completely different thing to actually implement it. Oh, and expect no 1-day software to support everything, and even less be bug-free! I rather wait months for a good driver rather than have experimental support for something, and then lose everything due to filesystem corruption.......
A closed mouth gathers no foot. Sky OS is not Linux, not a Linux distro or anything similar. Best not to comment if you have done no research.
http://www.skyos.org/?q=node/408
Oh and Linux does have DMA support.
Back in undergrad EE hardware labs I recall we did do something with DMA, maybe wrote some code for a 6802 and peripheral (maybe a pdp8), but then again it was pretty much all a replay of material put in front of us and of course the tutor was in the lab with any answers.
Since then I've done hardware ASICs with DMA support and the software guys still had a hard time of it even with us helping as much as possible. Its not just a matter of knowing where the control bits are but understanding somewhat how the hardware interacts with everything.
Anybody facing a motherboard and chipset and expecting to get much done with it quickly is going to be dissapointed, I bet you can find very little hardware docs for any chip at all other than really dated stuff or under NDA. For that reason I prefer hardware that does as little as possible but as fast as possible and let the software do most of it but the industry is doing the exact opposite. Its kind of amazing anyone else but MS gets anything done at all.
I also would have expected DMA before video graphics compositing but the latter is more interesting.
Using schelduler as counter example is totally pointless. You can have correct performance for desktop computing with a stupid scheduler. O(1) scheduler has an interest when you're running many concurrent process (waiting in queue, not locked in IO or sleeping). For servers with huge load, then. On the desktop, processor is idle at least 80% of the time, scheduler overhead might reduce reponseness, but not performance.
DMA is needed for performance, since modern hardware does almost all its data transfert through DMA. Locking CPU to, for example, write data to cache instead of rendering your web page will slow down your system much more than a O(n) scheduler with n = 4 on worst case. But i'am sure it would be still the case with n = 100.
I'm pretty surprised to see SkyOS implement DMA after a window compistor, for example.





