on Sat, Jun 07, 2003 at 07:34:34AM -0700, Paul Johnson (baloo@ursine.ca) wrote:
> On Sat, Jun 07, 2003 at 03:43:47PM +0200, aradorlinux@yahoo.es wrote:
> > > Would there be much of a speed increase, enough to warrent doing it ?
> >
> > Yes.
> >
> > (You can listen people saying "no"; but those people can't prove
> > how its posible that a optimized compilation of some apps seems to make
> > a diference in the real world)
>
> Minutes wasted for milliseconds gained. Still seem worth it?
While I'm not arguing for one conclusion or the other (and I don't
recompile my own apps), this is a simplistic answer. The thing is: it
depends on the minutes (and/or milliseconds).
As an analogy, PG&E in California maintains paired resevoirs in the Sierra
Nevada and coastal ranges. These are two resevoirs, connected by pipes
or tunnels, with a generating system in between them. Which doubles as
a pumping station.
Run one way, the stations generate electricity (water flows downhill).
Run the other way, the stations generate kinetic potential (water flows
uphill). Clearly, you can't generate more power than you expend pumping
water uphill (frictional, turbulent, and efficiency losses). The
difference is that peak load power is expensive (you are paying for
megawatt capacity, not megawatt-hours enerty, per se), and off-peak
power is cheap (even in pre-Enron days, by an order of magnitude).
Similarly, for sufficiently time-sensitive operations, even a relatively
intensive off-peak investment for a few seconds' boost on demand can
benefit. Or have you cancelled your nightly slocatedb build as well?
Realtime systems might be one example. Even in interactive use, various
levels of latency -- a few 100ths of a second for typing, less for
gameplaying or graphics where feedback control is significant, and 1-5
seconds for interactive lookups and queries -- matters to the computing
experience.
The unfortunate fact is that various interactive demands on system
performance are far more likely to effect perceived responsiveness
(swap, disk contention, etc.) than even the most rampant binary
optimizations. With latency, once you've started losing ground, you can
never get it back.
IIRC Jakob Nielsen and Josh Polsky have some interesting things to say
about this. There may also still be a copy of Stuart Cheshire's "It's
the Latency, Stupid" accessible from somewhere.
http://rescomp.stanford.edu/~cheshire/rants/Latency.html
Or similar:
http://peripherals.about.com/library/weekly/aa012101c.htm
Peace.
--
Karsten M. Self <kmself@ix.netcom.com> http://kmself.home.netcom.com/
What Part of "Gestalt" don't you understand?
"Yes," said Marvin. "Why stop now just when I'm hating it?"
-- HHGTG
Attachment:
pgpYtgF9arF1m.pgp
Description: PGP signature