Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2021-05-21
10:10:32 sean-k-mooney all data too and from the vm passes through br-int
10:10:45 sean-k-mooney regarless of if its vm to vm on same host or vxlan or vlan
10:10:45 sean-k-mooney regarless of if its vm to vm on same host or vxlan or vlan
10:11:03 sean-k-mooney so if we model the inventory there it models the datalane capsity
10:11:03 sean-k-mooney so if we model the inventory there it models the datalane capsity
10:11:12 sean-k-mooney by puttign the other bridge under it
10:11:13 sean-k-mooney by puttign the other bridge under it
10:11:21 sean-k-mooney we can model the actully bridge toplogy
10:11:21 sean-k-mooney we can model the actully bridge toplogy
10:11:28 sean-k-mooney and make same subtree work
10:11:28 sean-k-mooney and make same subtree work
10:11:39 bauzas ah
10:11:49 sean-k-mooney if you want 3 ovs instaces just create a second tree
10:11:49 sean-k-mooney if you want 3 ovs instaces just create a second tree
10:11:57 sean-k-mooney fo the second set of bridges and resrouses
10:11:57 sean-k-mooney fo the second set of bridges and resrouses
10:12:02 gibi (still it is technically equivalent to rename agent RP to br-int today)
10:12:02 gibi (still it is technically equivalent to rename agent RP to br-int today)
10:12:12 sean-k-mooney gibi: yes it is
10:12:12 sean-k-mooney gibi: yes it is
10:12:18 bauzas br-int is a neutron notion
10:12:18 bauzas br-int is a neutron notion
10:12:25 bauzas (well, a nova-net too but...)
10:12:25 bauzas (well, a nova-net too but...)
10:12:40 sean-k-mooney its the default name of the bridge we plug all the vms into
10:12:40 sean-k-mooney its the default name of the bridge we plug all the vms into
10:12:59 sean-k-mooney its also the name used with ovn and odl for what ites workth but its technically configurable
10:12:59 bauzas the neutron-agent seems to me more understandable from a x-project perspective
10:12:59 sean-k-mooney its also the name used with ovn and odl for what ites workth but its technically configurable
10:12:59 bauzas the neutron-agent seems to me more understandable from a x-project perspective
10:13:18 sean-k-mooney ack
10:13:18 sean-k-mooney ack
10:13:34 sean-k-mooney i said it was a slight crazy idea :)
10:13:34 sean-k-mooney i said it was a slight crazy idea :)
10:13:58 gibi interestingly we don't have a similar structure for the sriov side. There we have independent PFs managed by the same agent
10:13:58 gibi interestingly we don't have a similar structure for the sriov side. There we have independent PFs managed by the same agent
10:14:31 sean-k-mooney we could if we moded indevuate pci cards
10:14:31 sean-k-mooney we could if we moded indevuate pci cards
10:15:01 sean-k-mooney eg grouped PF together if they were on the same add in card but that has limited usefullness
10:15:01 sean-k-mooney eg grouped PF together if they were on the same add in card but that has limited usefullness
10:15:17 sean-k-mooney the main one woudl be pcie bandwith really
10:15:17 sean-k-mooney the main one woudl be pcie bandwith really
10:15:27 bauzas did I open a can of worms ?
10:15:27 bauzas did I open a can of worms ?
10:15:35 sean-k-mooney no
10:15:36 sean-k-mooney no
10:15:57 sean-k-mooney well its netwrokign so yes by default
10:15:57 sean-k-mooney well its netwrokign so yes by default
10:16:17 bauzas lol
10:16:17 bauzas lol
10:16:37 bauzas just trying to wrap up things in my mind
10:16:37 bauzas just trying to wrap up things in my mind
10:17:19 sean-k-mooney perhaps a call would also help?
10:17:19 sean-k-mooney perhaps a call would also help?
10:17:42 bauzas with the minimum bandwidth model, we said "heh, we have an agent which has bridges for inter-node communication' so we are going to put resources to track on those bridges
10:17:42 bauzas with the minimum bandwidth model, we said "heh, we have an agent which has bridges for inter-node communication' so we are going to put resources to track on those bridges
10:18:31 bauzas correct ?
10:18:31 bauzas correct ?
10:18:32 sean-k-mooney yes mainly because the bandwith is an aspect of a nic which is attached to at most 1 of thoes briges
10:18:32 sean-k-mooney yes mainly because the bandwith is an aspect of a nic which is attached to at most 1 of thoes briges
10:18:39 gibi yepp
10:19:00 sean-k-mooney the standard name for the bridge with the nice is generally br-ex
10:19:00 sean-k-mooney the standard name for the bridge with the nice is generally br-ex
10:19:01 bauzas so, local vm-to-vm communication thru br-int isn't guaranteed, right?
10:19:02 bauzas so, local vm-to-vm communication thru br-int isn't guaranteed, right?
10:19:16 openstackgerrit Brin Zhang proposed openstack/nova master: Replace all_tenants with all_projects in List Server APIs https://review.opendev.org/c/openstack/nova/+/765311
10:19:16 openstackgerrit Brin Zhang proposed openstack/nova master: Replace all_tenants with all_projects in List Server APIs https://review.opendev.org/c/openstack/nova/+/765311
10:19:17 openstackgerrit Brin Zhang proposed openstack/nova master: Replaces tenant_id with project_id from Rebuild Server API https://review.opendev.org/c/openstack/nova/+/766380
10:19:17 openstackgerrit Brin Zhang proposed openstack/nova master: Replaces tenant_id with project_id from Rebuild Server API https://review.opendev.org/c/openstack/nova/+/766380
10:19:28 sean-k-mooney bauzas: it does not consume any nic bandeith for kernel ovs
10:19:28 sean-k-mooney bauzas: it does not consume any nic bandeith for kernel ovs
10:19:42 sean-k-mooney bauzas: ovs's internal bandwith generally exceeed your nic bandwith
10:19:42 bauzas sean-k-mooney: I know, I just want to make things clear
10:19:42 sean-k-mooney bauzas: ovs's internal bandwith generally exceeed your nic bandwith
10:19:42 bauzas sean-k-mooney: I know, I just want to make things clear
10:19:47 sean-k-mooney but technically no its not
10:19:47 sean-k-mooney but technically no its not
10:20:06 sean-k-mooney we are jsut reserving inter host bandwith
10:20:06 sean-k-mooney we are jsut reserving inter host bandwith
10:20:09 bauzas cool
10:20:09 bauzas cool
10:20:14 bauzas here, back to the pps spec
10:20:15 bauzas here, back to the pps spec
10:20:30 bauzas (and gosh, we should find an acronym for the minimum bandwidth spec)
10:20:31 bauzas (and gosh, we should find an acronym for the minimum bandwidth spec)
10:20:38 gibi bw spec :)
10:20:38 gibi bw spec :)
10:20:40 bauzas min b/w, sorted it
10:20:40 bauzas min b/w, sorted it
10:20:51 bauzas ok, so, back to the pps spec
10:20:51 bauzas ok, so, back to the pps spec
10:20:51 sean-k-mooney :)
10:21:03 sean-k-mooney brb continue and ill catch up
10:21:03 sean-k-mooney brb continue and ill catch up
10:21:28 bauzas here, we want to guarantee the process rate of the internal ovs switch that's on... br-int ?
10:21:28 bauzas here, we want to guarantee the process rate of the internal ovs switch that's on... br-int ?
10:21:38 gibi * gibi free until the top of the hour, then need to jump to two consecutive calls 
10:21:56 bauzas * bauzas had to skip his usual Friday gym session :p
10:25:29 gibi in theory I we can rename the OVS Agent RP to br-int. It would help with some mental modelling. But this can be done any time independently of my work. So I would like to push this work to be a burden of the person who proposes the multiple OVS per agent spec. You can hate me for it, but I don't believe we ever get there. So I spare some effort not to do someting that will not be needed.
10:25:29 gibi in theory I we can rename the OVS Agent RP to br-int. It would help with some mental modelling. But this can be done any time independently of my work. So I would like to push this work to be a burden of the person who proposes the multiple OVS per agent spec. You can hate me for it, but I don't believe we ever get there. So I spare some effort not to do someting that will not be needed.

Earlier   Later