On 4/9/2020 2:38 AM, Alex Goodwin wrote:
>
>
> Thomas,
>
> So the hash and equality speedups I did for traveller-pyroute didn't
> really change that?
>
> On reflection, I suppose only reducing run time by 40-odd percent
> wouldn't do overly much.
>
> What sort of speedup would be needed to make the "submit custom gubbins
> and rerun for charted space" approach work?  Roughly a thousandfold?
>
> I got the impression (from reading GT:FT and traveller-pyroute numerous
> times) that the algorithm was inherently serial - for those of us
> alongside me in the cheap seats, I believe that it really couldn't be
> divided up and farmed out to multiple CPU cores to speed it up.
>
	The work you did was a great help. For a single sector the time to generate the
map and data is under 60 seconds, which is probably enough for a web interface.
But the whole of charted space is still running ~24 hours, so yes a thousand
fold performance.

	The python code is single-threaded. Python's ability to multi-thread compute
bound tasks is so limited it may as well be non-existent. To get the parallelism
it would need to be re-written in language more suited to it like Rust or Go.
Which I may consider at some point because these are the new hot languages.

	The other challenge, or what I consider a challenge, is adding a web interface.
The option of pasting in your own sector or selecting one from the Traveller Map
and having it process it. At that point I'm looking at wanting to have the whole
process in JavaScript and doing the work on the user's machine.

>
> Kaladorn,
>
> At least with the GT gravity-trade model and the travellerrpg data,
> long-range trade is _massive_ and interdependent.  From my own
> experiments (which gave rise to the patches I mentioned above), a
> single-sector trade map, standing alone, has vastly different trade
> patterns than that same sector in a full-Imperium or full-charted-space
> context.
>
> One experiment I tried resulted in side effects visible throughout
> Charted Space, with side orders of multiple, distinct, trade routes to
> systems within the Great Rift.  That temporarily diverted me into
> thinking about how to implement stable, usable wormholes in
> traveller-pyroute and their effects on trade.
>

	As part of one experiment for the Milieu 1900 I implemented having a Hop drive
(a J-10 drive with no shorter jump option). I'm quite sure adding in a set of
stable wormholes as an option wouldn't be that hard.

--
         Thomas Jones-Low
Work:	xxxxxx@softstart.com
Home:   xxxxxx@gmail.com