-
Bug
-
Resolution: Unresolved
-
Critical
-
None
-
4.14
-
Critical
-
None
-
False
-
-
Customer Escalated
Description of problem: The customer has an application that is intended to work using a Single thread & hence they tested using iPerf for bandwidth before deploying the application. The drop is significant as compared to HostNetwork (50-70 %). They are using a LACP-bonded configuration with two slave-interfaces included(25 Gbit/s NICs each) and a VLAN configured on top of it.
Version-Release number of selected component (if applicable): OVN-K with IPSec disabled
How reproducible:
Happens every time when used without parallel connections.
Steps to Reproduce:
1. Run iPerf to check pod-to-pod communication
2. Compare the results with & without HostNetwork: True
3. See if the packet drops are seen on the default interface.
Actual results:
#Hostnetwork: true Vs # Hostnetwork: false
Highest: 23.43 Gbit/s Highest: 10.43 Gbit/s
Average: 19.52 Gbit/s Average: 6.48 Gbit/s
Lowest: 10.2 Gbit/s Lowest: 3.12 Gbit/s
Expected results:
Drop is expected as OVN-K uses tunneling along with Genieve but this one seems more than 50% which is questionable
Additional info:
Already suggested them to increase RingBuffer as I had seen increased RX-DRP. I was hoping if there was any thing that could be tuned from the OVN side as well to help & hence raised this Slack thread to discuss.