Hi Everyone, I have been running performance tests and have some questions on the observations made in that process, especially with the role of vpp_main thread.
1. Noticed good performance when running vpp_main and vpp_worker_0 (polling on two 10G ports) on the same physical core (different logical cores). When vpp_worker_0 was moved to a different physical core altogether, started observing rx_queue tail drops. What is the relation of having vpp_main and vpp_worker_0 on the same physical core ? Why would moving a worker (which is processing the frames) to different core cause rx_queue tail drops ? 1. The device has two 10G and two 1G ports. Having 1G ports rx-queue polled by a different worker, vpp_worker_1 and keeping vpp_worker_0 polling on 10G ports gives good performance. When 1G ports are moved such that its rx queue’s is polled by vpp_main, there are rx_queue tail drops. I am looking to understand the relation between vpp_main and the workers ? Does vpp_main or any request from vpp_main thread stall the worker threads ? Thanks, Vijay
-=-=-=-=-=-=-=-=-=-=-=- Links: You receive all messages sent to this group. View/Reply Online (#12689): https://lists.fd.io/g/vpp-dev/message/12689 Mute This Topic: https://lists.fd.io/mt/30869592/21656 Group Owner: vpp-dev+ow...@lists.fd.io Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com] -=-=-=-=-=-=-=-=-=-=-=-
