On Fri, 2011-01-07 at 08:01 -0800, Cédric Beust ♔ wrote: [ . . . ] > > I don't see the logical implication between these two statements. > Cores are just a physical concept that software threads will benefit > from. The situation is not going to be radically different from today > where the idea of using threads to parallelize operations is quite > mainstream.
It is true that more cores per processor brings nothing that JVM-on-blade-system people haven't already seen -- i.e. hundreds of processors in a pool doing work sharing from a thread pool. However, there is a phase change. Instead of it being only available on big iron running single-thread-per-job (which side steps the whole concurrency parallelism problem) it is now available for everyone on every platform. Of course from HPC this is all a solved problem: you use Fortran or C++ with OpenMP and MPI. Go and D are trying to become competitive in this space by not competing: they are native code systems that push process and message passing as a computational model. So how will JVM compete in this "brave new world"? Given that future processors will be heterogeneous collections of computational units not just regular arrays of homogeneous, operating systems will have to change as will ideas of "application". It may be that no application will be written in a single language. Think about trying to write CUDA code targeting the JVM. If the processor contains GPGPUs and CPUs, then this has to be dealt with. > I really don't see why it's necessary to tell everyone "soon you'll > have much more cores on your machine, therefore you need to change the > way you program". Because these same people also think that shared-memory multi-threading is a good thing for all applications development? The fact that they have got away with not really knowing how to work with threads properly has been a fluke. -- Russel. ============================================================================= Dr Russel Winder t: +44 20 7585 2200 voip: sip:russel.win...@ekiga.net 41 Buckmaster Road m: +44 7770 465 077 xmpp: rus...@russel.org.uk London SW11 1EN, UK w: www.russel.org.uk skype: russel_winder
signature.asc
Description: This is a digitally signed message part