Hi All, The below mentioned performance issue with goes away, when we removed the following two lines from API "linux_epoll_input_inline" in the file "vpp/src/src/vlib/unix/input.c"
if (timeout_ms) usleep (timeout_ms * 1000); Please let us know the your opinion on the changes. Thanks, Padma R From: Padmavathy Ramakrishnan Sent: Thursday, September 5, 2019 10:58 PM To: 'vpp-dev@lists.fd.io' <vpp-dev@lists.fd.io> Cc: 'vpp-dev@lists.fd.io' <vpp-dev@lists.fd.io> Subject: RE: throughput degradation observed when threads are not assigned to any queues/interfaces Hi All, Is there any update on this issue? Please let us if you need more inputs from our side Thanks, Padma R From: Padmavathy Ramakrishnan Sent: Friday, August 30, 2019 1:24 PM To: vpp-dev@lists.fd.io<mailto:vpp-dev@lists.fd.io> Subject: throughput degradation observed when threads are not assigned to any queues/interfaces Hi All, we are observing performance issues ( less throughput) when we have one free thread in VPP, the thread (vpp_wk_2 ) is not assigned with any interface or queue. vpp# show threads ID Name Type LWP Sched Policy (Priority) lcore Core Socket State 0 vpp_main 18887 other (0) 1 0 0 1 vpp_wk_0 workers 18889 other (0) 10 5 0 2 vpp_wk_1 workers 18890 other (0) 12 6 0 3 vpp_wk_2 workers 18891 other (0) 14 7 0 in the below output vpp_wk_2 is not assigned with any queue/interface. vpp# show interface rx-placement Thread 1 (vpp_wk_0): node dpdk-input: TenGigabitEthernet6/0/1 queue 0 (polling) Thread 2 (vpp_wk_1): node memif-input: memif502/1000 queue 0 (polling) vpp# performance issue goes away when we remove the free core from startup.conf or when we assign at least one queue/interface (though it is not receiving any traffic) to the free thread vpp_wk_2. For example: in the below output interface memif503/1002 is assigned to vpp_wk_2. memif503/1002 is not used for frame processing. vpp# show interface rx-placement Thread 1 (vpp_wk_0): node dpdk-input: TenGigabitEthernet6/0/1 queue 0 (polling) Thread 2 (vpp_wk_1): node memif-input: memif502/1000 queue 0 (polling) Thread 3 (vpp_wk_2): node memif-input: memif503/1002 queue 0 (polling) vpp# Please let us know why the performance is going down when we have one free thread/core in the VPP. Thanks, Padma R
-=-=-=-=-=-=-=-=-=-=-=- Links: You receive all messages sent to this group. View/Reply Online (#13942): https://lists.fd.io/g/vpp-dev/message/13942 Mute This Topic: https://lists.fd.io/mt/33170876/21656 Group Owner: vpp-dev+ow...@lists.fd.io Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com] -=-=-=-=-=-=-=-=-=-=-=-
