home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.programming
- Path: sparky!uunet!gatech!hubcap!mjs
- From: mjs@hubcap.clemson.edu (M. J. Saltzman)
- Subject: Re: C specific editor (was Re: C/C++ Speed)
- Message-ID: <1993Jan21.193158.3853@hubcap.clemson.edu>
- Organization: Clemson University, Clemson SC
- References: <1993Jan17.041622.5602@thunder.mcrcim.mcgill.edu> <47740001@hplvec.LVLD.HP.COM>
- Date: Thu, 21 Jan 1993 19:31:58 GMT
- Lines: 52
-
- In article <47740001@hplvec.LVLD.HP.COM> elw@hplvec.LVLD.HP.COM (Eddie Williamson) writes:
- >----------
- >>> In article <1993Jan15.160435.19703@acuson.com>,
- >>> miklg@acuson.com (Michael Goldman ) writes:
- >
- >>> The whole discussion of speed is limited to inexperienced people. In
- >>> real world applications, the hardware will double your speed in the
- >>> time it takes to write the application.
- >
- >----------
- >> hplvec:comp.programming
- >> kurt@tc.fluke.COM (Kurt Guntheroth)
- >> 9:54 am Jan 18, 1993 /
- >>
- >> The net result is that in the *real* real world, efficiency is *always*
- >> important. In the synthetic real world imagined by acedemics, you can wave
- >> your magic wand and say "just buy enough hardware to make it fast", but when
- >> you are spending your own money, or money that comes out of your profit
- >> sharing bonus, or from somebody other than nameless taxpayers, you have to
- >> buy what you can afford, and live with what you buy.
- >----------
- >I agree! (with the second paragraph)
- >Most real world customers can't affort to go out and buy the latest
- >hardware every time they buy new application software. They live with
- >what they already have. Don't make excuses for being lazy. Design your
- >software from the beginning with efficiency in mind. You'll end up
-
- As an academic and a government employee, I'd just like to point out
- that we don't get to spend nameless taxpayers' money just to go out
- and buy new hardware every time we buy new software, any more than
- anyone else. In fact, we often don't even get to go out and buy the
- software! We also buy what we can afford and live with what we buy,
- and we often live longer with less than many in the so-called real
- world.
-
- Algorithm designers don't abstract the constant factor away because
- they think you can always make it up by buying new hardware--they do
- it because they want to make a hardware-independent comparison of
- algorithms. On the other hand, it usually really is the case that
- better asymptotic performance beats a constant-factor reduction (if
- the constant isn't ridiculous to begin with), especially when it's
- important (when problems get large).
-
- If it's important to solve moderate-sized problems quickly, then the
- constant *is* important. But it's also often true that most of the
- constant factor can be had with attention to only a small portion of
- the code. You can almost always pick this up after the implementation
- is complete with careful profiling.
- --
- Matthew Saltzman
- Clemson University Math Sciences
- mjs@clemson.edu
-