[PD] parallel processing

[Andy Farnell]

It's funny reading old predictions that didn't quite work out (yet).
One nuclear power station supplying all the planet with 3 giant
supercomputers (presumably running with less than 640k or RAM) managing
all human problems while we live a life of leisure and swan about in 
our flying cars...

Yeah. Right.

We were also taught something in computer science classes back in 88
that unlike Moores Law and other popularised concepts isn't quite 
so widely talked about. If it were 

a) these predictions wouldn't seem so riduculous
b) people wouldn't be so quick to make them

It's called the "wheel of life". If you could see it, technological
development might look like a toroidal vortex. Things migrate away
from the main CPU to become independent subsystems as they mature
and specialise. Then they are subsumed back into the main area (be it
motherboard vs peripheral card or specialised instruction sets on
co-processors)

Then the cycle repeats.

You can see it in everything, DRAM and DMA controllers, sound 
and video chips/cards, maths co-processors, network controllers.

We are already on the second rotation in sound. Once there were
special sound chips that used hybrid AD synthesis, like the SID.

Then it went native.

Then it went to DSPs.

Then DSPs were obsoleted by cost and lack of flexibilty
and it went native again.

No doubt we will see another turn giving us massively parallel
specialised SPU (sound/signal processors) that can run 100 instances
of Pd or something, before that too gets folded back in to the
silicon as an integrated faculty. 

A new influence on the block is green/environmental considerations.
It's no good having power hungry specialised subsystems running idle
for most of the time if it can be done natively.

I don't want to embarrass the authors, because we all make bold
judgements that come back to haunt us (and that's
a good thing to venture an opinion and risk being wrong rather than
have nothing to say), but I've read plenty of similar comments 
that trumpet the amazing parallel flexibility of dataflow 
programming. Of course this misses the point that writing parallel
programs requires analysis and algorithm development with that in 
mind, it isn't just a bonus you get for free when you have more than
one processor.

IRCAMS experimental multi-processor synthesisers and things like
the Kyma and MARS were stepping stones along the way. Lessons
learned can be incorporated into new implementations of Pd-like
dataflow languages. The good thing is the establishment of a
language and method that has potential to hide implementation.
Presumably Pd would not look very different to the programmer
if it were to have parallel scheduling and would scale seamlessly
from one to many processors.











On Tue, 04 Mar 2008 11:14:25 -0500
marius schebella <marius.schebella at gmail.com> wrote:

> hi,
> I am reading an old interview with james moorer (with curtis roads in 
> CMJ/6 1982). one funny thing is that he says, 'software synthesis is 
> either dead or dying[...] I am hoping it's demise will be quick and 
> relatively painless.'
> in return he predicted all computation being done on special dsp chips. 
> in part he was right, but on the other hand the main cpu got more than 
> fast enough to survive (gfx is slightly different), but - and I am 
> coming to my point - he also was thinking about hundreds or thousands of 
> parallel processing elements. right now, we are going to have several 
> and in the future many many parallel CPUs, and the need for parallel 
> processing is back. miller was talking about that in montreal.
> so I wonder how pd will survive that evolution? afaik the current 
> situation is poor in this regard. can anyone give an outview for the 
> future? would it be a jump from pd (I) 0.43 to pd II 0.1?
> marius.
> 
> _______________________________________________
> PD-list at iem.at mailing list
> UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list


-- 
Use the source