perf(jetsocat,dgw): rework flow control algorithm #977
Conversation
This file contains 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
Replace the old naive flow control algorithm. The new algorithm is much more fair, and considerably reduce the number of WindowAdjust messages exchanged. The idea is to send WindowAdjust messages only when the scheduler is idle, and focus on handling active connections as fast as possible without sending any WindowAdjust message when it is busy. Before this patch, single connection: > 0.0000-15.5887 sec 34.4 GBytes 18.9 Gbits/sec After this patch + maximum message size of 4 kB, single connection: > 0.0000-10.0490 sec 23.0 GBytes 19.7 Gbits/sec Improvement of 4.2%. This patch + maximum message size of 16 kB, single connection: > 0.0000-10.0393 sec 30.6 GBytes 26.2 Gbits/sec Improvement of 38.6%. Before this patch, 10 parallel connections: > [ ID] Interval Transfer Bandwidth > [ 3] 0.0000-20.0677 sec 555 MBytes 232 Mbits/sec > [ 9] 0.0000-20.0704 sec 3.89 GBytes 1.67 Gbits/sec > [ 2] 0.0000-20.0703 sec 3.62 GBytes 1.55 Gbits/sec > [ 4] 0.0000-20.0693 sec 3.91 GBytes 1.67 Gbits/sec > [ 10] 0.0000-20.0659 sec 3.43 GBytes 1.47 Gbits/sec > [ 5] 0.0000-20.0651 sec 2.73 GBytes 1.17 Gbits/sec > [ 8] 0.0000-20.0704 sec 3.83 GBytes 1.64 Gbits/sec > [ 7] 0.0000-20.0692 sec 3.37 GBytes 1.44 Gbits/sec > [ 1] 0.0000-20.0692 sec 3.61 GBytes 1.54 Gbits/sec > [ 6] 0.0000-20.2801 sec 455 MBytes 188 Mbits/sec > [SUM] 0.0000-20.2810 sec 29.4 GBytes 12.4 Gbits/sec Notice how some connections are considerably slower. After this patch + maximum message size of 4 kB, 10 parallel connections: > [ 8] 0.0000-10.3368 sec 1.88 GBytes 1.56 Gbits/sec > [ 10] 0.0000-10.3377 sec 1.87 GBytes 1.55 Gbits/sec > [ 1] 0.0000-10.3642 sec 1.87 GBytes 1.55 Gbits/sec > [ 5] 0.0000-10.3640 sec 1.86 GBytes 1.54 Gbits/sec > [ 2] 0.0000-10.3640 sec 1.85 GBytes 1.54 Gbits/sec > [ 4] 0.0000-10.3650 sec 1.85 GBytes 1.54 Gbits/sec > [ 3] 0.0000-10.3641 sec 1.85 GBytes 1.54 Gbits/sec > [ 9] 0.0000-10.3646 sec 1.90 GBytes 1.58 Gbits/sec > [ 6] 0.0000-10.3652 sec 1.85 GBytes 1.54 Gbits/sec > [ 7] 0.0000-10.3800 sec 1.87 GBytes 1.55 Gbits/sec > [SUM] 0.0000-10.3812 sec 18.7 GBytes 15.4 Gbits/sec Improvement of 24.1%. Also notice how all connections are transferring roughly the same amount of bytes.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Replace the old naive flow control algorithm.
The new algorithm is much more fair, and considerably reduce the number of WindowAdjust messages exchanged.
The idea is to send WindowAdjust messages only when the scheduler is idle, and focus on handling active connections as fast as possible without sending any WindowAdjust message when it is busy.
Before this patch, single connection:
After this patch + maximum message size of 4 kB, single connection:
Improvement of 4.2%.
This patch + maximum message size of 16 kB, single connection:
Improvement of 38.6%.
Before this patch, 10 parallel connections:
Notice how some connections are considerably slower.
After this patch + maximum message size of 4 kB, 10 parallel connections:
Improvement of 24.1%.
Also notice how all connections are transferring roughly the same amount of bytes.
Issue: DGW-202