Hi, We have a heavy node that does lots of packet field modifications (like encap tunnels) and is the next node for dpdk-input. This node is followed by another node and outputs the packet to the dpdk-output node. We are observing some packet corruption in case we have more than 1 interface configured to this core. If there is a 1 to 1 mapping between the cores and the interfaces then we do not see this corruption.
We have a some questions regarding the vector processing in this case. 1. Is it that each interface generates its own vector ? 2. How does the nodes get processed ? Is it that the first node goes through all the vectors and then schedules the next node or 3. is it that a vector is taken up for processing and it completes all the node processing in the graph before picking up the next vector ? 4. We have a requirement to pass some context data between the nodes. The opaque data is unfortunately small to hold these information. What is the recommendation in general to pass around this context information ? Regards, Vinay
-=-=-=-=-=-=-=-=-=-=-=- Links: You receive all messages sent to this group. View/Reply Online (#10205): https://lists.fd.io/g/vpp-dev/message/10205 Mute This Topic: https://lists.fd.io/mt/24612433/21656 Group Owner: vpp-dev+ow...@lists.fd.io Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com] -=-=-=-=-=-=-=-=-=-=-=-
