Reduce dataplane overhead caused by ct action #3858
Merged
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
|
/test-all |
|
This should increase Windows dataplane performance as well since CT action is the main factor affecting the performance. |
|
Adding it to milestone 1.7 since in increases dataplane performance and may simplify flows #3603 will add. |
Codecov Report
@@ Coverage Diff @@
## main #3858 +/- ##
==========================================
- Coverage 64.36% 57.31% -7.05%
==========================================
Files 288 403 +115
Lines 41239 57057 +15818
==========================================
+ Hits 26542 32703 +6161
- Misses 12557 21740 +9183
- Partials 2140 2614 +474
Flags with carried forward coverage won't be shown. Click here to find out more.
|
hongliangl
previously approved these changes
Jun 2, 2022
jianjuns
previously approved these changes
Jun 2, 2022
antoninbas
previously approved these changes
Jun 2, 2022
pkg/agent/openflow/framework.go
Outdated
| @@ -219,14 +219,14 @@ func (f *featureService) getRequiredTables() []*Table { | |||
| return []*Table{DNATTable} | |||
| } | |||
| tables := []*Table{ | |||
| SNATConntrackTable, | |||
| UNSNATTable, | |||
There was a problem hiding this comment.
nit: I think UnSNAT instead of UNSNAT makes more sense and is more readable
To support performing both SNAT and DNAT for traffic, Antrea uses two CT zones for SNAT and DNAT separately. For each packet, multiple CT actions are executed to go through the zones. And because SNAT is performed after DNAT, reply traffic wouldn't be unNATed correctly if they go through the zones in the same order as request traffic, an extra CT action for unSNAT was added before DNAT to resolve it. These CT actions introduce measurable overhead to the dataplane. Since the first unSNAT action is for reply traffic of SNATed connections only, and there are only few cases needing SNAT, this patch adds conditions to the unSNAT flow to make irrelevant traffic bypass it. With less CT action and less recirculation caused by it, the dataplane performance is significantly increased. TCP_RR and TCP_CRR improvement in a kind cluster is as below: ``` Test old TPS new TPS delta TCP_RR 14568.69 17826.26 +22.36% TCP_CRR 2781.7 3498.12 +25.75% ``` Signed-off-by: Quan Tian <qtian@vmware.com>
tnqn
dismissed stale reviews from antoninbas, jianjuns, hongliangl, and wenyingd
via
04af829
|
/test-all |
hongliangl
approved these changes
Jun 5, 2022
jianjuns
approved these changes
Jun 5, 2022
|
/test-networkpolicy |
tnqn
added
the
action/release-note
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.
To support performing both SNAT and DNAT for traffic, Antrea uses two CT zones for SNAT and DNAT separately. For each packet, multiple CT actions are executed to go through the zones. And because SNAT is performed after DNAT, reply traffic wouldn't be unNATed correctly if they go through the zones in the same order as request traffic, an extra CT action for unSNAT was added before DNAT to resolve it. These CT actions introduce measurable overhead to the dataplane.
Since the first unSNAT action is for reply traffic of SNATed connections only, and there are only few cases needing SNAT, this patch adds conditions to the unSNAT flow to make irrelevant traffic bypass it.
With less CT action and less recirculation caused by it, the dataplane performance is significantly increased. TCP_RR and TCP_CRR improvement in a kind cluster is as below:
Signed-off-by: Quan Tian qtian@vmware.com