By: Gabriele Svelto (gabriele.svelto.delete@this.gmail.com), December 8, 2014 7:32 pm
Room: Moderated Discussions
Jouni Osmala (josmala.delete@this.cc.hut.fi) on December 8, 2014 1:10 pm wrote:
> I'm assuming that 90+% of programs already run fast enough and they don't matter for this.
In today's mobile world "fast enough" might not be the end of it all as far as optimization goes. Since most computing devices now run on batteries making something that's fast enough for the user perception faster is actually an important goal: it allows you to save power. In this respect certain parallel methods provide a bad trade-off: all else being equal they can improve the overall speed of execution at the cost of increased computation (and often communication) over an equivalent serial approach. Now in practice not everything outside of computation is equal so you've got fixed costs to amortize too and overall faster execution might be more efficient nevertheless; but this just to point out that there's more variables to take into account nowadays when deciding if a workload needs to be sped up or not.
> I'm assuming that 90+% of programs already run fast enough and they don't matter for this.
In today's mobile world "fast enough" might not be the end of it all as far as optimization goes. Since most computing devices now run on batteries making something that's fast enough for the user perception faster is actually an important goal: it allows you to save power. In this respect certain parallel methods provide a bad trade-off: all else being equal they can improve the overall speed of execution at the cost of increased computation (and often communication) over an equivalent serial approach. Now in practice not everything outside of computation is equal so you've got fixed costs to amortize too and overall faster execution might be more efficient nevertheless; but this just to point out that there's more variables to take into account nowadays when deciding if a workload needs to be sped up or not.
