On 11/13/06, Devdas Bhagat <[EMAIL PROTECTED]> wrote:
On 13/11/06 12:26 +0100, Dave Long wrote: <snip> > (from a 50 year old document)[1] > >There are occasions when it is necessary to write programs which will > >be executed in as little time as possible. The programmer should bear > >in mind that 10,000 executions of all non-optimum instructions would > >take less than 3 minutes longer than 10,000 executions of optimum > >instructions. If the programmer spends 15-30 minutes on each routine > >to save machine time by optimizing, this time may never be made up in > >the actual running of the problem. > The converse is also true. If you are writing code which will be used by a large number of people, saving a few cycles on every run will save a lot of cycles overall. For examples of applications which have programmers working on features rather than performance, see MS Windows, MS Office, Open Office, GNOME.
I can't find reference to it so maybe this is folklore, but my favourite example is Jobs' appeal to engineers to cut the boot time on the original mac. They'd worked their arses off to make impressive progress on boot time and said no more could be done. He asked them to do the maths on saving one second, assuming 1 million customers and a near daily boot for three years. They asked for a little more time to try and come up with more ideas to shave off another few seconds.
Devdas Bhagat
