[generator] Performance regression? #6737

New issue

Open

opened 2023-11-30 15:35:43 +00:00 by pastk · 0 comments

pastk commented

2023-11-30 15:35:43 +00:00

Member

My local map generation times increased significantly for some cases after #6537

E.g. Luxembourg generation time increased x2.7 (from 16min to 43min).

The biggest time consumer is in-between following log lines:
(note a 2243s difference)

LOG TID(1) INFO      142.471 generator/search_index_builder.cpp:694 BuildSearchIndex(): End building search index, elapsed seconds: 6.81536
LOG TID(1) INFO      142.496 generator/search_index_builder.cpp:617 BuildSearchIndexFromDataFile(): Search index size = 3475111
LOG TID(1) INFO      2385.41 generator/search_index_builder.cpp:582 BuildAddressTable(): Saved streets entries number: 108166
LOG TID(1) INFO      2385.41 generator/search_index_builder.cpp:583 BuildAddressTable(): Saved places entries number: 626
LOG TID(1) INFO      2385.41 generator/search_index_builder.cpp:588 BuildAddressTable(): Matched addresses percent: 99.7881 Total: 109022 Missing: 231
LOG TID(1) INFO      2385.44 generator/search_index_builder.cpp:626 BuildSearchIndexFromDataFile(): Streets table size: 239447 ; Places table size: 88932

Before the PR it was 591s (so this step consumes x3.8 more time)

LOG TID(1) INFO      138.762 generator/search_index_builder.cpp:615 BuildSearchIndex(): End building search index, elapsed seconds: 6.42105
LOG TID(1) INFO      138.787 generator/search_index_builder.cpp:539 BuildSearchIndexFromDataFile(): Search index size = 3219382
LOG TID(1) INFO      729.683 generator/search_index_builder.cpp:507 BuildAddressTable(): Address: BuildingToStreet entries count: 108130
LOG TID(1) INFO      729.683 generator/search_index_builder.cpp:513 BuildAddressTable(): Address: Matched percent 99.9519 Total: 108182 Missing: 52
LOG TID(1) INFO      729.719 generator/search_index_builder.cpp:546 BuildSearchIndexFromDataFile(): Search address table size = 239159

But e.g. for Montenegro the difference is smaller: the total generation time increased just 10%, while the above mentioned step has increased by x3.

Is this performance decrease justified or it could be optimized?

FYI @vng

My local map generation times increased significantly for some cases after #6537 E.g. Luxembourg generation time increased x2.7 (from 16min to 43min). The biggest time consumer is in-between following log lines: (note a 2243s difference) ``` LOG TID(1) INFO 142.471 generator/search_index_builder.cpp:694 BuildSearchIndex(): End building search index, elapsed seconds: 6.81536 LOG TID(1) INFO 142.496 generator/search_index_builder.cpp:617 BuildSearchIndexFromDataFile(): Search index size = 3475111 LOG TID(1) INFO 2385.41 generator/search_index_builder.cpp:582 BuildAddressTable(): Saved streets entries number: 108166 LOG TID(1) INFO 2385.41 generator/search_index_builder.cpp:583 BuildAddressTable(): Saved places entries number: 626 LOG TID(1) INFO 2385.41 generator/search_index_builder.cpp:588 BuildAddressTable(): Matched addresses percent: 99.7881 Total: 109022 Missing: 231 LOG TID(1) INFO 2385.44 generator/search_index_builder.cpp:626 BuildSearchIndexFromDataFile(): Streets table size: 239447 ; Places table size: 88932 ``` Before the PR it was 591s (so this step consumes x3.8 more time) ``` LOG TID(1) INFO 138.762 generator/search_index_builder.cpp:615 BuildSearchIndex(): End building search index, elapsed seconds: 6.42105 LOG TID(1) INFO 138.787 generator/search_index_builder.cpp:539 BuildSearchIndexFromDataFile(): Search index size = 3219382 LOG TID(1) INFO 729.683 generator/search_index_builder.cpp:507 BuildAddressTable(): Address: BuildingToStreet entries count: 108130 LOG TID(1) INFO 729.683 generator/search_index_builder.cpp:513 BuildAddressTable(): Address: Matched percent 99.9519 Total: 108182 Missing: 52 LOG TID(1) INFO 729.719 generator/search_index_builder.cpp:546 BuildSearchIndexFromDataFile(): Search address table size = 239159 ``` But e.g. for Montenegro the difference is smaller: the total generation time increased just 10%, while the above mentioned step has increased by x3. Is this performance decrease justified or it could be optimized? FYI @vng

This repo is archived. You cannot comment on issues.