Re: Are these all really true ?

>> * Memory is free, speed is what is worth optimizing.
>
>Actually, your algorithms are what is worth optimising. That way,
>you might get more speed, and still use less memory. After all,
>if you're running on a multi-user system, and you use just too
>much memory, your process will spent most of its time in page faults.


I agree with the need for picking a good algorithm. But if processors
and memory continue to advance at their current rates, then in the
next year or two we're going to start seeing systems where cache
misses can cost you on the order of 500 instructions. Just as (or
maybe more) important than page faults are cache misses, and the
"memory is free" attitude can lead you down the wrong path quickly.


I think people are going to have to start dusting off their old 70's
algorithms books that discussed sorting/searching on disk/tape in
order to rekindle the thinking processes used to come up with
algorithms that run well on tomorrow's systems.


(Of course multi-threaded CPUs get you around this (are you out there,
Preston? :-))


So, I not only disagree with the "memory is free" statement, I think
that that attitude is going to hurt people who *are* optimizing for
speed.


(BTW: Don't forget the TLB. Whether or not your data resides on the
number of pages you can map with the TLB can be just as important as
whether or not you've squeezed the data into the number of bytes you
can put in your cache)


-----------------------------------------------------------------------------
Jay Finger (finger@convex.com) CONVEX Computer Corporation
Kernel Development Engineer P.O. Box 833851, 3000 Waterview Pkwy
All opinions are my own Richardson, TX 75083-3851
--

Return to the comp.compilers page.
Search the comp.compilers archives again.