Why Physicists Still Use Fortran
Meta
jared771 at gmail.com
Tue Oct 17 16:31:07 UTC 2017
On Tuesday, 17 October 2017 at 13:09:37 UTC, Steven Schveighoffer
wrote:
> Ouch! I had an experience like that once.
>
> I worked at a company that bought a one-man show's company who
> had an impressive load-balancing software we wanted to
> incorporate in our system.
>
> About 1-2 years into him working at our company, one of our
> developers tested it using webbench (all testing had been done
> by this guy previously), and was getting terrible numbers. But
> his tests always showed really good numbers.
>
> Turns out he was "timing" his benchmarks by starting a separate
> thread, then sleeping for 1 second, and then measuring how many
> requests he handled in that "1 second". But of course, the
> system was super-loaded, so the sleep was going way longer than
> 1 second, and his numbers looked great! After we fixed it, the
> numbers looked horrific and matched webbench.
>
> When this was found out, we kind of moved away from that
> software, as we were moving our focus to hardware. I can't
> imagine how that must have felt, though.
>
> -Steve
This is just plain negligence on upper management's part. I can't
believe they got that far without doing due diligence to verify
his results.
More information about the Digitalmars-d
mailing list