Thank you for your reply, comments inline. On Jun 5, 2011, at 4:25 PM, Dale Schumacher <[email protected]> wrote:
> On Sun, Jun 5, 2011 at 5:23 PM, Casey Ransberger > <[email protected]> wrote: >> >> Has anyone taken the actor model down to the metal? > > If someone has, I would sure like to hear about it! There was the > Apiary machine, but I don't think that was ever physically built, only > simulated. Googling... > (snip) > The SEND and BECOME primitives seem fairly straight-forward to > translate to hardware. It is the CREATE primitive that I struggle > with. > Since we can't actually "create" new hardware elements (snip) Oh, yeah. That makes sense. > Maybe there would be some way to activate latent nodes of processing > power, injecting them with their initial behavior as a way of > breathing life into them. I really like this idea. > It could be just a matter of "allocating" > new actors the way we allocate memory. Each hardware node could have > a capacity of available actors who only need a script to become alive. This is not far off from what I was already daydreaming about. When I started I thought those guys looked like a kind of regular "object animator" that would light up when something was bound. I'd likely have to cache the ones that didn't fit on the chip somewhere. Maybe to deal with concurrency I should really start thinking of them as "actor animators". I'm sure there's a way to pull this off. Even if it's by having a lot of FPGAs on the logic board so that I can compensate for reconfiguration latency by switching between them, but I don't think that idea fits any goal around a parsimonious architecture, which is one thing that I'm after. The synchronization problems I'd expect also seem awful, unless someone out there has thought a bunch about doing a low-level TeaTime (or what have you.) So I'm really hoping I can find a general thing that I can just place many identical copies of in the "die" or whatever it is we use now... ahem. I am such a noob! And then just swap them out to main memory or a local cache when I run out. > I would love to explore this idea further and hear how you would > consider approaching the problem. I will definitely CC you if I think I've gotten somewhere with it. Feel free to send me a note if you have any big aha-moments, because I have a tiny slab of time to run at that fence before I'm going to have to get back to work, and any/all help that I can get will be much appreciated. If I made it, I'd likely build a couple of boxes and try to pass them off as art (like what one buys for the wall,) but my plan is to make everything you need ("IP cores" appears to be the term of industry) to do it yourself available under the MIT license if and when I've made some actual progress. I reckon I have a better shot at getting to actually use it if I just give it away! _______________________________________________ fonc mailing list [email protected] http://vpri.org/mailman/listinfo/fonc
