Closed kevindweb closed 5 years ago
I purposely turned off CI in testing, I will definitely run CI before merge.
Won't merge before CI approves this though 😛
Resolves issue #62 I think? Not 64
@onvm will you be kind?
@onvm will you be kind?
Your results will arrive shortly
@onvm you're next
@onvm you're next
Another CI run in progress, adding request to the end of the list
@onvm you're next
Your results will arrive shortly
qqq
Your results will arrive shortly
qqq
Error: ERROR: Script failed on nimbnode30
Testing
Your results will arrive shortly
Referencing issue #62 , this pr adds a macro
FLOW_LOOKUP
toonvm_pkt.c
to enable/disable our flow table lookup.Summary:
Although this is a "good first issue", it's been an issue for a while, and @koolzz and I found it was very useful for testing the pktgen functionality in CI. Here's what I received with this added functionality. With
ENABLE_FLOW_LOOKUP=1
, from the max rx was13331727
, max tx was14811688
. Disabling the flow table lookup portion ofonvm_pkt_process_rx_batch
, we received max rx of14783964
, and max tx of14785566
. Over 1mil pps rx faster by disabling the flow table, which makes sense, but is good that we got it.Usage: By default, per @twood02 's request in the original issue, flow lookups happen. But in
/onvm/onvm_mgr/Makefile
settingENABLE_FLOW_LOOKUP=0
, flow lookups will not happen. This is useful if only one port is running pktgen.Merging notes:
I want to merge this into the upcoming CI pktgen update, to leverage this functionality. We saw a 6mil pps increase by using this on the cluster nodes the other day. (no strict dependencies though)
TODO before merging :
Test Plan:
Make sure pktgen still works.
Review:
@koolzz you were there on Saturday talking about this @dennisafa I know you work a lot with Pktgen, if you could test this