Tuesday, 2025-11-04

*** mhen_ is now known as mhen02:53
Renich[m]Hello. I am looking on information on how to deploy a DCN using kolla-ansible. I'd appreciate any pointers. 09:44
luchaaijer[m]<Renich[m]> "Hello. I am looking on informati..." <- I do not have any experience with a DCN setup, but from the looks of it a multinode setup with one AZ per site would work.11:21
luchaaijer[m]Networking might be a bit finnicky to get right, but that depends on your setup11:21
luchaaijer[m]Do you have any specific requirements?11:22
*** sambork_ is now known as sambork12:11
Renich[m]For now, the idea is to have a control plane on virtual machines at GCP. Additionally, compute nodes on bare metal at datacenters.16:28
Renich[m][@luca:utwente.io](https://matrix.to/#/@luca:utwente.io) thank you for answering. :)16:28
JayFRenich[m]: why?18:13
JayFRenich[m]: that seems like a silly architecture, I'm curious what's the value in not self-hosting that control plane18:13
Renich[m][@_oftc_JayF:matrix.org](https://matrix.to/#/@_oftc_JayF:matrix.org) the intention is to have more than a few computes in several different datacenters18:18
JayFOH! I think I misread the original18:19
JayFI work on Ironic so when I read "compute nodes on bare metal at datacenters" I assumed Ironic18:19
JayFyou're just talking about Hypervisors18:19
JayFstill a bit strange to me not to self-host the control plane (openstack-on-openstack except in the same cloud, an oroboros architecture so to speak)18:20
JayFbut more understandable18:20
Renich[m]JayF: :D18:29
Renich[m]This is a company that wants to sell very specific services based on openstack on several datacenters. 18:40
Renich[m]The idea of a centralized controlled plane is just to simplify the architecture and be able to manage everything from a central location. 18:40
Renich[m]I know it is not the common openstack paradigm. We are looking for options. 18:40
Renich[m]I think this could be considered edge computing or sorts.18:40
Renich[m]I checked out StarlingX and Airship as well.18:41
JayFah, that makes more sense18:45
JayFlike you run and host everything but the nova-computes, which live in the datacenter18:46
JayFI would be very curious how well that would perform with the latency to/from the queue18:46
Renich[m]Indeed. We made a test using wireguard and it wasn't terrible... with a single user running an instance.18:47
JayFI know in Ironic, we have folks who will often run a central control plane to manage remote hardware18:50
JayFin those cases, we usually see folks draw the line between Ironic API and Conductor, having disparate conductor groups in different regions18:50
JayFI'm unsure the shape of similar features in other projects, but I imagine there's some help there18:50

Generated by irclog2html.py 4.0.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!