Dima calculation times xDima2 (HPR)

[Bernd Welter]

Hi there,

here is an overview of some sample computation based on a 40'000 test set...
Calculation times grow in square:

Size HPR Dima	Calculation times in [sec]	size in [MB]	equi CONV	Speedup
(500)²	0,897	1,5
(1'000)²	2	6
(2'500)²	9,1	37,5	25² = 11 sec
(5'000)²	20	150	50² = 16 sec	(5000²) / (50²) = 10'000
(7'500)²	41	337,5
(10'000)²	65= 1:05	600	100² = 60 sec
(15'000)²	142 = 2:22	1350	200² = 180 sec
(20'000)²	240 = 4:00	2400
(25'000)²	365 = 6:05	3750	300² = 384 sec	(25'000)²/(300)² = 7'000
(30'000)²	507 = 8:27	5400	400² = 620 sec
(35'000)²	710 = 11:50	7350
(40'000)²	945 = 15:45	9600	500² = 900 sec	(40'000)²/(500)² = 6'400

Calculated on a DELL Precision 5530, Intel i7-8850H (2.6 GHz), 32.0GB, Windows 10-64bit

Best regards,
Bernd

• Update 3.11.2023: I added the column "equi CONV" to the table. It contains some comparison value that shows "what dima size can you compute in the same time via CONVENTIONAL routing". The base is my current laptop (Dell Intel I7-11850H / 64GB) and the GER PLZ5 codes / car profile / xServer v2.30.

[Bernd Welter]

Quick update: I have a new laptop:

• Processor: 11th Gen Intel(R) Core(TM) i7-11850H @ 2.50GHz 2.50 GHz
• Memory: 64,0 GB (63,7 GB usable)

I computed a very large (57'xxx) matrix = with almost 3.5 billion relations! in less than 20 minutes.

Size o the disk: 26 GB

[Bernd Welter]

and another mega distance matrix based on 62'434 pairwise different (x,y) computed with an HPR based on
- car, no further truck attributes
- geographic restriction: allowed countries = ["IT"]
took me 10:48 minutes and occupies 29GB memory for it's 3'898'004'356 relations...

[Bernd Welter]

Wow... I've just been told that one of my partners managed to calculate a (103'000)² matrix within 68 minutes...
That's very fast for 10 Billions of relations!

On the other hand I ased him why he requires such a huge data set... probably such a "surrounding" problem can be simplified with some preaggregations ...