Jakob Praher <[EMAIL PROTECTED]> wrote: > > Am Fre, 2003-02-28 um 17.48 schrieb Hunsberger, Peter: > > Pier Fumagalli wrote: > > > > > > "Hunsberger, Peter" <[EMAIL PROTECTED]> wrote: > > > > > > > Niclas Hedhman wrote: > > > >> > > > >> I'm playing with the idea of making an attempt at a > SAX hardware > > > >> accelerator, just for the heck of it, but won't bother > if it is > > > >> already in the "ignorable" > > > >> magnitude. > > > > > > > > You mean something like: > > > > > > > > http://www.datapower.com/products/xa35.html > > > > > > > > Of course, it doesn't run Cocoon, but if you've really > got a heavy > > > > duty XSLT load it may be something that you can justify > > > wiring up on a > > > > custom basis... > > > > > > Interesting... XSLTC but done native instead of done through > > > a JVM... I wonder if they generate straight optimized > > > bytecode for the processor directly interpreting the stylesheet... > > > > > > Pier (old assembly fart) > > > > > > <snip/>
> how would you use such a tool? > If it is designed as network node, which consumes xml, parses > it, transforms it and then sends it back somewhere, the > question is how much you gain, since the fact that the > network bus speed is tremendously slow, compared to bus > speeds within a system. > Just guessing but I believe it's designed as a network edge component; taking in incoming XML and transforming it for your apps and taking outgoing XML and transforming it for your partners. In effect I think it's real purpose is XML format conversion, which is a bit of a niche, but I think we're all starting to see how non-standard "standard" XML formats are... I don't think it takes much to extend this niche to general purpose transformation; replacing app severs for Web servers and perhaps even replacing Web servers, but then getting the data to the beast becomes an issue, though presumably you can still beat the WAN speed with your LAN... > It would be interesting to have an active component within a > computer, like those server optimized network cards, that > takes some computing cycles from the cpu by doing xml/xslt > based work - but that is probably hard. You need some low level hooks, but that's what video cards do all the time: move transformations from the CPU to the video card. I'd guess this is where Niclas was heading...??? > > and more problems come in when you have to use some kind of > custom exstension library... . > Makes it more interesting, but it's still like a video card: have it hardware render what it can, do the rest yourself. > anyway very interesting read, > > thanks > > Niclas, Peter, Pier >
