On 3/27/2012 12:23 PM, Miles Fidelman wrote:
karl ramberg wrote:
Slides/pdf:
http://www.dynamic-languages-symposium.org/dls-11/program/media/Ungar_2011_EverythingYouKnowAboutParallelProgrammingIsWrongAWildScreedAboutTheFuture_Dls.pdf
Granted that their approach to an OLAP cube is new, but the folks
behind Erlang, and Carl Hewitt have been talking about massive
concurrency, an assuming inconsistency as the future trend, for years.
Miles Fidelman
yeah, it seems a bit like an overly dramatic and hand-wavy way of saying
stuff that probably many of us knew already (besides maybe some guy off
in the corner being like "but I thought mutex-based locking could scale
up forever?!...").
granted, language-design may still need some work to find an ideal
programming model for working with concurrent systems, but I still more
suspect it will probably end up looking more like "existing language
with better concurrency features bolted on" than "some fundamentally new
approach to programming" (like, say, C or C++ or C# or ActionScript or
similar, with message-passing and constraints or similar bolted on).
another issue I can think of:
how does Tilera compare, say, with AMD Fusion?...
a quick skim over what information I could find was not showing any
strong reasons (technical or economic) which would be leaning in
Tilera's favor vs AMD Fusion (maybe there is something more subtle?...).
both seem to be highly parallel VLIW architectures, ...
granted, as-is, one is still stuck using things like CUDA or OpenCL, but
maybe something can be found to be able to largely eliminate needing
these (or, gloss over them).
a partial idea that comes up is that of having a sort of bytecode format
which can be compiled into the particular ISAs of the particular cores.
or, alternatively, throwing some sort of "x86 to VLIW JIT /
trans-compiler" or similar into the mix.
or such...
_______________________________________________
fonc mailing list
fonc@vpri.org
http://vpri.org/mailman/listinfo/fonc