| opendevreview | Dmitriy Rabotyagov proposed openstack/openstack-ansible-os_nova master: Rename nova_enabled_vgpu_types https://review.opendev.org/c/openstack/openstack-ansible-os_nova/+/816466 | 11:49 |
|---|---|---|
| opendevreview | Dmitriy Rabotyagov proposed openstack/openstack-ansible stable/ussuri: Fetch upper constraints file with curl rather than allow pip to download it https://review.opendev.org/c/openstack/openstack-ansible/+/815632 | 12:18 |
| opendevreview | Dmitriy Rabotyagov proposed openstack/ansible-role-python_venv_build stable/train: Workaround distro provided pip having old CA certs on centos-7 https://review.opendev.org/c/openstack/ansible-role-python_venv_build/+/816473 | 12:19 |
| opendevreview | Dmitriy Rabotyagov proposed openstack/openstack-ansible stable/train: Fetch upper constraints file with curl rather than allow pip to download it https://review.opendev.org/c/openstack/openstack-ansible/+/815633 | 12:19 |
| opendevreview | Merged openstack/ansible-hardening master: Explicitly create clamav socket directory https://review.opendev.org/c/openstack/ansible-hardening/+/814748 | 12:57 |
| opendevreview | James Gibson proposed openstack/ansible-role-pki master: Force regen of cert when requested https://review.opendev.org/c/openstack/ansible-role-pki/+/816520 | 14:30 |
| opendevreview | Merged openstack/openstack-ansible-ceph_client master: Ensure role not fail when mon_host is not part of ceph_extra_confs https://review.opendev.org/c/openstack/openstack-ansible-ceph_client/+/814586 | 14:36 |
| opendevreview | James Gibson proposed openstack/ansible-role-pki master: Force regen of cert when requested https://review.opendev.org/c/openstack/ansible-role-pki/+/816520 | 14:39 |
| opendevreview | James Gibson proposed openstack/openstack-ansible master: Add playbook to generate any user defined certificates https://review.opendev.org/c/openstack/openstack-ansible/+/816522 | 14:49 |
| spatel | jamesdenton quick question | 15:56 |
| spatel | https://paste.opendev.org/show/810364/ | 15:56 |
| spatel | why br-int and all other bridges showing (tap) ? | 15:56 |
| jamesdenton | good question. i would think they would be 'internal' | 16:11 |
| opendevreview | Dmitriy Rabotyagov proposed openstack/ansible-role-systemd_service master: Add ability to create templated services https://review.opendev.org/c/openstack/ansible-role-systemd_service/+/816531 | 16:19 |
| opendevreview | Dmitriy Rabotyagov proposed openstack/ansible-role-systemd_service master: Add ability to create templated services https://review.opendev.org/c/openstack/ansible-role-systemd_service/+/816531 | 16:20 |
| opendevreview | James Gibson proposed openstack/openstack-ansible-os_nova master: Enable TLS for live migrations https://review.opendev.org/c/openstack/openstack-ansible-os_nova/+/815224 | 16:26 |
| mgariepy | spatel, did you found the tap vs internal stuff ? | 18:38 |
| spatel | mgariepy tap vs internal? | 18:41 |
| mgariepy | <spatel> https://paste.opendev.org/show/810364/ | 18:41 |
| spatel | oh jamesdenton replied | 18:41 |
| spatel | that is interesting in my case they are tap so don't understand why? | 18:42 |
| mgariepy | i'm on internal also with ovs. | 18:42 |
| spatel | may be that is my issue for performance | 18:42 |
| mgariepy | but i don't have dpdk. | 18:42 |
| mgariepy | how are your network interface configured? | 18:42 |
| spatel | OVS creating these bridges | 18:42 |
| mgariepy | you pass it a vlan interface direcly ? | 18:43 |
| mgariepy | or an untag interface | 18:43 |
| spatel | Yes. i have vlan based provider | 18:43 |
| spatel | VLAN | 18:43 |
| spatel | now i am really curious who created tap there? | 18:44 |
| spatel | who decided that? | 18:45 |
| mgariepy | and it's like that accross all nodes? or it's only one that you debugged stuff on it ? | 18:45 |
| spatel | all my DPDK compute nodes showing same output | 18:45 |
| spatel | can you show me your output ? | 18:47 |
| spatel | holy crap! | 18:48 |
| jamesdenton | here's mine: https://paste.opendev.org/show/810367/ | 18:49 |
| spatel | I have one more deployment which has this output but non-DPDK - https://paste.opendev.org/show/810368/ | 18:49 |
| spatel | This is OVN based deployment | 18:49 |
| jamesdenton | same | 18:49 |
| spatel | does DPDK changing this to tap? | 18:49 |
| spatel | or OSA + DPDK | 18:50 |
| jamesdenton | it could be the difference between non-dpdk and dpdk ovs? | 18:50 |
| spatel | but that is bad.. tap interface is kernel space and it will hurt performance | 18:50 |
| spatel | or may be when we switch to netdev it will replace internal with tap. hmmm | 18:51 |
| jamesdenton | yeah but that interface is the port, not the bridge | 18:51 |
| jamesdenton | so likely unrelated | 18:51 |
| spatel | trying to think why tap instead of internal when doing dpdk | 18:53 |
| jamesdenton | "Open vSwitch has a network device type called “tap”. This is intended only for implementing “internal” ports in the OVS userspace switch and should not be used otherwise. In particular, users should not configure KVM “tap” devices as type “tap” (use type “system”, the default, instead)." | 18:55 |
| jamesdenton | maybe related | 18:55 |
| jamesdenton | https://docs.openvswitch.org/en/latest/faq/issues/ | 18:56 |
| spatel | I talked to one of INTEL employee who is working on DPDK and i gave my configuration to him and he said, this is wrong. he told me tap interface is bottleneck and he said create VM and attach to vhost_user directly without any OVS etc.. and run performance you will see 15mpps with DPDK | 18:57 |
| spatel | so what is the verdict here, this is normal? | 18:59 |
| jamesdenton | normal? not sure. i hesitate to say it's related to your issues | 19:01 |
| dmsimard | FYI, upstream EOL of ansible 2.9 and ansible-base 2.10 have been announced: https://groups.google.com/g/ansible-announce/c/kegIH5_okmg/ | 19:03 |
| spatel | i am seeing throughput of 3gbps on vm which is almost related to tap throughput, now i am going to run one more test which is create dpdk port and bind it directly to VM using vhost_user and then run loadtest, if i get good result then my issue is somewhere in ovs-dpdk | 19:05 |
| spatel | feel like OVS components are bottleneck | 19:06 |
Generated by irclog2html.py 2.17.2 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!