Ah, yes, I didn't mean to say that things are impossible, a good GC is already an example of a useful but leaky abstraction with appropriate knobs. Another example is a database like datomic. I mean that these tradeoffs and how they change with greater numbers of cores will force our hands in one way or another, and we shouldn't be surprised when sticking to what's natural and useful on the small-scale doesn't work out on the large-scale.
It might make things like one jvm per core + IPC worthwhile when it seems silly now. On Sun, Apr 28, 2013 at 1:18 PM, Andy Fingerhut <andy.finger...@gmail.com>wrote: > If the issue is that the GC implementation in Oracle's JVMs typically has > a sequential bottleneck in it with default parameters, then yes, Amdahl's > law is striking *for that GC implementation*. > > For problems that are embarassingly parallel (little or no communication > or shared memory between threads), there is no law of the universe that > says that GC must have a sequential bottleneck in it. A GC better tuned > for such problems could be developed, and perhaps already has. > > Lee, did you ever get a trial for Azul's Zing JVM to see if its GC was any > better in this regard? > > Andy > > > On Sun, Apr 28, 2013 at 8:47 AM, Gary Trakhman <gary.trakh...@gmail.com>wrote: > >> Amdahl strikes again! I knew this lunch was too tasty to be free. >> >> I just thought of another analogy. When I think of something highly >> concurrent and scalable on the JVM, I think of web request handlers, which >> effectively use no memory-sharing for domain-specific work, at all, unless >> there is some in-memory application state, but in general they shift the >> problem to a database. When the bookkeeping overhead doesn't dominate, >> they appear to be highly concurrent. >> >> Shared-memory a wall of abstraction, multi-core cpus don't work this way >> at the bottom level, so writing programs this way can only go so far. >> >> >> >> >> On Sun, Apr 28, 2013 at 11:22 AM, Manuel Paccagnella < >> manuel.paccagne...@gmail.com> wrote: >> >>> Ok, for everyone that is interested: >>> this<http://www.meetup.com/Functional-Programming-Connoisseurs/messages/boards/thread/30946382>thread >>> on meetup clarified things a bit for me and probably it's a good >>> summary on the specific case that Lee found. >>> >>> >>> Il giorno domenica 28 aprile 2013 17:07:04 UTC+2, Lee ha scritto: >>> >>>> >>>> On Apr 28, 2013, at 10:57 AM, Manuel Paccagnella wrote: >>>> > >>>> > This is an interesting perspective. Are you saying that currently >>>> Clojure doesn't offer adequate tools to take full advantage of parallel >>>> execution on multi-core machines? >>>> >>>> I am speaking only from my own experience, and only relative to my own >>>> case. But I've been unable to get reasonable speedups for my system, which >>>> I think should be an excellent candidate for multi-core speedups. And while >>>> I'm not an expert on JVM/Clojure performance I solicited and received >>>> advice from people on this list who I think are, and I'm still unable to >>>> get reasonable speedups -- maybe a couple of X but maybe not even that, on >>>> machines with core-counts up to 48. It may have to do with the kinds of >>>> tasks I'm running (lots of sequence manipulation and memory consumption, >>>> etc.), but for whatever reason I am still unable to get Clojure's >>>> oft-touted multi-core benefits. >>>> >>>> -Lee >>>> >>>> -- >>> -- >>> You received this message because you are subscribed to the Google >>> Groups "Clojure" group. >>> To post to this group, send email to clojure@googlegroups.com >>> Note that posts from new members are moderated - please be patient with >>> your first post. >>> To unsubscribe from this group, send email to >>> clojure+unsubscr...@googlegroups.com >>> For more options, visit this group at >>> http://groups.google.com/group/clojure?hl=en >>> --- >>> You received this message because you are subscribed to the Google >>> Groups "Clojure" group. >>> To unsubscribe from this group and stop receiving emails from it, send >>> an email to clojure+unsubscr...@googlegroups.com. >>> For more options, visit https://groups.google.com/groups/opt_out. >>> >>> >>> >> >> -- >> -- >> You received this message because you are subscribed to the Google >> Groups "Clojure" group. >> To post to this group, send email to clojure@googlegroups.com >> Note that posts from new members are moderated - please be patient with >> your first post. >> To unsubscribe from this group, send email to >> clojure+unsubscr...@googlegroups.com >> For more options, visit this group at >> http://groups.google.com/group/clojure?hl=en >> --- >> You received this message because you are subscribed to the Google Groups >> "Clojure" group. >> To unsubscribe from this group and stop receiving emails from it, send an >> email to clojure+unsubscr...@googlegroups.com. >> For more options, visit https://groups.google.com/groups/opt_out. >> >> >> > > -- > -- > You received this message because you are subscribed to the Google > Groups "Clojure" group. > To post to this group, send email to clojure@googlegroups.com > Note that posts from new members are moderated - please be patient with > your first post. > To unsubscribe from this group, send email to > clojure+unsubscr...@googlegroups.com > For more options, visit this group at > http://groups.google.com/group/clojure?hl=en > --- > You received this message because you are subscribed to the Google Groups > "Clojure" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to clojure+unsubscr...@googlegroups.com. > For more options, visit https://groups.google.com/groups/opt_out. > > > -- -- You received this message because you are subscribed to the Google Groups "Clojure" group. To post to this group, send email to clojure@googlegroups.com Note that posts from new members are moderated - please be patient with your first post. To unsubscribe from this group, send email to clojure+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/clojure?hl=en --- You received this message because you are subscribed to the Google Groups "Clojure" group. To unsubscribe from this group and stop receiving emails from it, send an email to clojure+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/groups/opt_out.
