[Libre-soc-dev] gigabit router design decisions

[lkcl]

On November 4, 2021 9:42:17 PM UTC, Jacob Lifshay <programmerjake at gmail.com> wrote:

>Sorry that it sounded like you are incompetent, that is not at all what
>I
>meant. imho *no one*, not just you, is all that good at estimating how
>long
>programming tasks take, they can easily take waay less or waay more
>time
>than estimated.

i learned - again, 20 years experience - that drilling down (splitting) to tasks of approximately 1/2 to 5 days, and splitting into many many tasks (5 minumum, 30 is better) gives an averaging effect that resulted consistently in around 15% accuracy, despite many of those tasks being +/- 30% on their minimum and maximum estimate.  (poisson distribution effect or something going on, there)

(unfortunately it would often take about 2-3 days to get that level of detail, and i wouldn't get paid at the end of it, but hey)

it requires actual prior experience of actually doing those tasks before you can guage them properly.

guessing (and not putting in margins) or worse not having done something and missing out steps, *this* is how you get disastrously bad estimates.

by having been the one to do the HDL, documentation, unit tests, bugfixing, layout, everything, i *do* know how many tasks are needed, and *do* know how long they are likely to take (+/- 30%)

(where i didn't before, and got several things wrong)

this is now our second ASIC, not the first.

i know what needs to be done.

l.