Collection of minor performance optimizations #2855
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
… modulus with FastMod 9aac9f9 replace modulus with FastMod (Martin Ankerl) Pull request description: Not sure if this is optimization is necessary, but anyway I have some spare time so here it is. This replaces the slow modulo operation with a much faster 64bit multiplication & shift. This works when the hash is uniformly distributed between 0 and 2^32-1. This speeds up the benchmark by a factor of about 1.3: ``` RollingBloom, 5, 1500000, 3.73733, 4.97569e-07, 4.99002e-07, 4.98372e-07 # before RollingBloom, 5, 1500000, 2.86842, 3.81630e-07, 3.83730e-07, 3.82473e-07 # FastMod ``` Be aware that this changes the internal data of the filter, so this should probably not be used for CBloomFilter because of interoperability problems. Tree-SHA512: 04104f3fb09f56c9d14458a6aad919aeb0a5af944e8ee6a31f00e93c753e22004648c1cd65bf36752b6addec528d19fb665c27b955ce1666a85a928e17afa47a
In one of my profiling sessions with many InstantSend transactions happening, calls into CSporkManager added up to about 1% of total CPU time. This is easily avoidable by using unordered maps.
Due to the batched pruning, there is no need to maintain an ordered map of values anymore. Only when nPruneAfterSize, there is a need to create a temporary ordered vector of values to figure out what can be removed.
… on demand CNode::AskFor will now push entries into an initially unordered vector instead of an ordered multimap. Only when we later want to use vecAskFor in SendMessages, we sort the vector. The vector will actually be mostly sorted in most cases as insertion order usually mimics the desired ordering. Only the last few entries might need some shuffling around. Doing the sort on-demand should be less wasteful then trying to maintain correct order all the time.
UdjinM6
previously approved these changes
Apr 11, 2019
UdjinM6
left a comment
There was a problem hiding this comment.
utACK
I'm surprised that tiny sporks and args map had such a noticeable impact tbh, sounds very strange to me.
Author
|
The sporks is a different thing. We now perform a lot of |
This ensures that future backports that depends on limitedmap's ordering conflict so that we are made aware of needed action.
Author
|
Added a few commits to fix Travis and also ensure that future backports don't go wrong. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
See individual commits. Most of these are related to using unordered maps/sets instead of ordered ones and are the result of multiple profiling sessions. All these small optimizations add up to about 10% of CPU usage on the message handler thread.
This PR also contains a backport of bitcoin#13176 as I've also noticed unexpected high CPU use for
CNode::AddInventoryKnown.