LB transit legs

[mority]

computes a lower bound for number of transit legs required for every station to reach the destination

wallclock times for some cross continent queries in transitous:

all stations reached after 12 rounds, 340ms
all stations reached after 13 rounds, 257ms
all stations reached after 12 rounds, 342ms
all stations reached after 14 rounds, 256ms
all stations reached after 12 rounds, 341ms
all stations reached after 14 rounds, 256ms
all stations reached after 14 rounds, 344ms
all stations reached after 12 rounds, 255ms

[mority]

I just remembered, we discussed an approach to possibly speed this up a while ago.

Once we reach a station S, we reach all other stations that are connected to S via a footpath with the same amount of transit legs. Hence, we can pack all stations connected via footpaths into components and create a graph, where the nodes are components and the edges indicate which components are reachable with a single transit leg.

Might be more compact and therefore faster?

[Kiarash-Parvizi]

Benchmark

• Compiler: GCC
• Dataset: Germany
• Algorithm: Pong

lb_transits lazily disabled for rraptor & ping phase

Metric	Value
Average	994.40
Max	8,980
P99	4,917
P90	2,559
P80	1,703
Median (P50)	516
Min	72

Baseline (master at 73ee556)

Metric	Value
Average	1,164.89
Max	9,940
P99	6,531
P90	3,180
P80	1,995
Median (P50)	548
Min	70

Delta vs. master

Metric	Change
Average	-14.6%
P99	-24.7%
P90	-19.5%
P80	-14.6%
Median (P50)	-5.8%