Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Optimise getNetworkDistance in MLD even more #6884

Merged
merged 20 commits into from
May 23, 2024

Conversation

SiarheiFedartsou
Copy link
Member

@SiarheiFedartsou SiarheiFedartsou commented May 18, 2024

Issue

The idea is to calculate distance during routing and get rid of separated loop after routing this way. I left some comments over PR to make it more clear. See match_mld benchmark below to see performance improvement.

Tasklist

Requirements / Relations

Link any requirements here. Other pull requests this PR is based on?

Benchmark Results

Benchmark Base PR
alias aliased u32: 1142.86
plain u32: 1143.9
aliased double: 1210.16
plain double: 1211.98
aliased u32: 1155.42
plain u32: 1138.74
aliased double: 1212.36
plain double: 1208.4
json-render String: 8.43291ms
Stringstream: 11.6403ms
Vector: 7.67532ms
String: 8.59148ms
Stringstream: 11.6016ms
Vector: 7.66364ms
match_ch Default radius:
4.46997ms/req at 82 coordinate
0.0545118ms/coordinate
Radius 5m:
4.46891ms/req at 82 coordinate
0.054499ms/coordinate
Radius 10m:
15.1349ms/req at 82 coordinate
0.184572ms/coordinate
Radius 15m:
36.815ms/req at 82 coordinate
0.448963ms/coordinate
Radius 30m:
313.327ms/req at 82 coordinate
3.82106ms/coordinate
Default radius:
4.47117ms/req at 82 coordinate
0.0545264ms/coordinate
Radius 5m:
4.4615ms/req at 82 coordinate
0.0544086ms/coordinate
Radius 10m:
15.2068ms/req at 82 coordinate
0.185449ms/coordinate
Radius 15m:
37.1068ms/req at 82 coordinate
0.452521ms/coordinate
Radius 30m:
316.303ms/req at 82 coordinate
3.85735ms/coordinate
match_mld Default radius:
3.37703ms/req at 82 coordinate
0.0411833ms/coordinate
Radius 5m:
3.42276ms/req at 82 coordinate
0.041741ms/coordinate
Radius 10m:
12.6636ms/req at 82 coordinate
0.154434ms/coordinate
Radius 15m:
31.4966ms/req at 82 coordinate
0.384105ms/coordinate
Radius 30m:
350.316ms/req at 82 coordinate
4.27215ms/coordinate
Default radius:
2.84541ms/req at 82 coordinate
0.0347001ms/coordinate
Radius 5m:
2.83189ms/req at 82 coordinate
0.0345352ms/coordinate
Radius 10m:
10.3248ms/req at 82 coordinate
0.125912ms/coordinate
Radius 15m:
26.4356ms/req at 82 coordinate
0.322385ms/coordinate
Radius 30m:
301.352ms/req at 82 coordinate
3.67502ms/coordinate
packedvector random write:
std::vector 11330.3 ms
util::packed_vector 82423.3 ms
slowdown: 7.27459
random read:
std::vector 11209.3 ms
util::packed_vector 33739.8 ms
slowdown: 3.00998
random write:
std::vector 11350.7 ms
util::packed_vector 82211 ms
slowdown: 7.24284
random read:
std::vector 11168.9 ms
util::packed_vector 33560.4 ms
slowdown: 3.00481
route_ch 1000 routes, 3 coordinates, no alternatives, overview=full, steps=true
594.941ms
0.594941ms/req
1000 routes, 2 coordinates, no alternatives, overview=full, steps=true
375.079ms
0.375079ms/req
1000 routes, 2 coordinates, 3 alternatives, overview=full, steps=true
760.43ms
0.76043ms/req
1000 routes, 3 coordinates, no alternatives, overview=false, steps=false
152.323ms
0.152323ms/req
1000 routes, 2 coordinates, no alternatives, overview=false, steps=false
99.1662ms
0.0991662ms/req
1000 routes, 2 coordinates, 3 alternatives, overview=false, steps=false
134.189ms
0.134189ms/req
1000 routes, 3 coordinates, no alternatives, overview=false, steps=false, radius=750
152.045ms
0.152045ms/req
1000 routes, 2 coordinates, no alternatives, overview=false, steps=false, radius=750
98.8639ms
0.0988639ms/req
1000 routes, 2 coordinates, 3 alternatives, overview=false, steps=false, radius=750
133.923ms
0.133923ms/req
1000 routes, 3 coordinates, no alternatives, overview=full, steps=true
600.409ms
0.600409ms/req
1000 routes, 2 coordinates, no alternatives, overview=full, steps=true
378.786ms
0.378786ms/req
1000 routes, 2 coordinates, 3 alternatives, overview=full, steps=true
769.029ms
0.769029ms/req
1000 routes, 3 coordinates, no alternatives, overview=false, steps=false
161.375ms
0.161375ms/req
1000 routes, 2 coordinates, no alternatives, overview=false, steps=false
103.56ms
0.10356ms/req
1000 routes, 2 coordinates, 3 alternatives, overview=false, steps=false
145.516ms
0.145516ms/req
1000 routes, 3 coordinates, no alternatives, overview=false, steps=false, radius=750
161.27ms
0.16127ms/req
1000 routes, 2 coordinates, no alternatives, overview=false, steps=false, radius=750
104.048ms
0.104048ms/req
1000 routes, 2 coordinates, 3 alternatives, overview=false, steps=false, radius=750
145.495ms
0.145495ms/req
route_mld 1000 routes, 3 coordinates, no alternatives, overview=full, steps=true
727.614ms
0.727614ms/req
1000 routes, 2 coordinates, no alternatives, overview=full, steps=true
464.97ms
0.46497ms/req
1000 routes, 2 coordinates, 3 alternatives, overview=full, steps=true
947.942ms
0.947942ms/req
1000 routes, 3 coordinates, no alternatives, overview=false, steps=false
279.794ms
0.279794ms/req
1000 routes, 2 coordinates, no alternatives, overview=false, steps=false
166.12ms
0.16612ms/req
1000 routes, 2 coordinates, 3 alternatives, overview=false, steps=false
298.845ms
0.298845ms/req
1000 routes, 3 coordinates, no alternatives, overview=false, steps=false, radius=750
273.348ms
0.273348ms/req
1000 routes, 2 coordinates, no alternatives, overview=false, steps=false, radius=750
164.025ms
0.164025ms/req
1000 routes, 2 coordinates, 3 alternatives, overview=false, steps=false, radius=750
295.475ms
0.295475ms/req
1000 routes, 3 coordinates, no alternatives, overview=full, steps=true
729.539ms
0.729539ms/req
1000 routes, 2 coordinates, no alternatives, overview=full, steps=true
471.886ms
0.471886ms/req
1000 routes, 2 coordinates, 3 alternatives, overview=full, steps=true
957.81ms
0.95781ms/req
1000 routes, 3 coordinates, no alternatives, overview=false, steps=false
284.501ms
0.284501ms/req
1000 routes, 2 coordinates, no alternatives, overview=false, steps=false
168.564ms
0.168564ms/req
1000 routes, 2 coordinates, 3 alternatives, overview=false, steps=false
303.357ms
0.303357ms/req
1000 routes, 3 coordinates, no alternatives, overview=false, steps=false, radius=750
277.989ms
0.277989ms/req
1000 routes, 2 coordinates, no alternatives, overview=false, steps=false, radius=750
167.514ms
0.167514ms/req
1000 routes, 2 coordinates, 3 alternatives, overview=false, steps=false, radius=750
303.569ms
0.303569ms/req
rtree 1 result:
201.022ms -> 0.0201022 ms/query
10 results:
237.758ms -> 0.0237758 ms/query
1 result:
206.792ms -> 0.0206792 ms/query
10 results:
237.055ms -> 0.0237055 ms/query

@SiarheiFedartsou SiarheiFedartsou changed the title Optimise getNetworkDistance even more Optimise getNetworkDistance in MLD even more May 18, 2024
@SiarheiFedartsou SiarheiFedartsou force-pushed the sf-optimise-getNetworkDistance branch from 0dee408 to 83f34af Compare May 20, 2024 09:49
@SiarheiFedartsou SiarheiFedartsou force-pushed the sf-optimise-getNetworkDistance branch from 5fa491f to 4cdfae0 Compare May 20, 2024 10:51
@SiarheiFedartsou SiarheiFedartsou force-pushed the sf-optimise-getNetworkDistance branch from 7701c5f to edc4522 Compare May 21, 2024 17:29
{
// Shortcuts in forward direction
const auto &cell =
cells.GetCell(metric, level, partition.GetCell(level, heapNode.node));
auto destination = cell.GetDestinationNodes().begin();
auto distance = [&cell, node = heapNode.node ]() -> auto
{
if constexpr (IS_MAP_MATCHING)
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We are using new if constexpr feature from C++17, so it allows to avoid any impact on existing code (if it is not map matching the whole compiler will remove branch and will leave only contents of else clause).

@@ -281,14 +301,31 @@ void relaxOutgoingEdges(const DataFacade<Algorithm> &facade,

const auto level = getNodeQueryLevel(partition, heapNode.node, args...);

static constexpr auto IS_MAP_MATCHING =
std::is_same_v<typename SearchEngineData<mld::Algorithm>::MapMatchingQueryHeap, Heap>;
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I added new MapMatchingQueryHeap to be able to track distance in each heap entry.

}

for (size_t index = 0; index < unpacked_nodes.size() - 1; ++index)
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The essence of optimisation: we are getting rid of this loop, and track distance during routing instead.

@@ -269,10 +270,29 @@ retrievePackedPathFromHeap(const SearchEngineData<Algorithm>::QueryHeap &forward
return packed_path;
}

template <bool DIRECTION, typename Algorithm, typename... Args>
template <typename Heap>
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

My main goal here was to make all these algorithms to track distance if they are used in map matching and avoid any overhead in other places: I eventually found a way to do that via using if constexpr

const auto toHeapNode = forward_heap.GetHeapNodeIfWasInserted(to);
if (!toHeapNode)

if constexpr (IS_MAP_MATCHING)
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Tbh all these copy-pasted ifs are a bit ugly, but it seems to be a price for no overhead...

@SiarheiFedartsou SiarheiFedartsou marked this pull request as ready for review May 22, 2024 18:51
@DennisOSRM
Copy link
Collaborator

Nice improvement. Well done. 👍🏽

@SiarheiFedartsou SiarheiFedartsou merged commit 21607e0 into master May 23, 2024
19 of 20 checks passed
@SiarheiFedartsou SiarheiFedartsou deleted the sf-optimise-getNetworkDistance branch May 23, 2024 18:46
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants