On 07 May 2011, at 02:06, Igor Stasenko wrote: > On 6 May 2011 23:45, Stefan Marr <[email protected]> wrote: >> There is nothing fundamental in the RoarVM that is changing the language >> semantics of Smalltalk. >> >> It is just that for: `[do something] fork` you will have to assume that it >> is executed in parallel to other code. >> > only that? > Heh.. then we're 99% done. Except that this last 1% is still could > take years to complete :)
But why? We show with the Squeak 3.x MVC image that it is perfectly possible. (the change set is absolutely minimal, and the only thing we needed to fix was some Delay related issues) And, on top of that it is perfectly possible, there is a smooth transition path. You can tell all your Process objects to just run on the main core in the beginning, and then just use the other cores for code you wrote specifically. Thus, there is a nice step-wise path. And I think I outlined that in my 20min presentation at the school after Stef asked. BTW: are those videos somewhere accessible? From my perspective all it takes is someone who actually cares, and has the time to experiment a bit with parallel programming. The kernel, and the tools can be migrated step by step. However, that is just on top of the current RoarVM, which is certainly not as attractive as a CogVM with real thread support. > I can tell you more: there is no business cases for VM(s) which can do > manycore :) No manycore perhaps, but luckily the low-power end is pushing to multicore solutions. So, your next iPhone and Android app will need to leverage at least two cores for optimal performance, and even quad-core phones are just around the corner. > Implementing VM which enabling massive parallelism is just a > beginning. Then obviously you need do to a lot at language side > to leverage that, in order to really say "yes, our system(s) are aware > of multicore and can scale almost linearly in future". No, don't think so. If people got deadlines and a need, they will do it with the tools at hand. And what I am doing here is absolutely not rocket science. Actually it is pretty hard to sell the engineering we do as science at all. Because the multicore VM problem has been solved a decade ago. They just forgot to tell us how... Best regards Stefan -- Stefan Marr Software Languages Lab Vrije Universiteit Brussel Pleinlaan 2 / B-1050 Brussels / Belgium http://soft.vub.ac.be/~smarr Phone: +32 2 629 2974 Fax: +32 2 629 3525
