Initial comments as PTL inline…
From: <[email protected]> on behalf of Jim Baker <[email protected]> Date: Thursday, August 13, 2020 at 7:44 AM To: Jack Morgan <[email protected]>, "STEINBRUECK, SCOTT A" <[email protected]> Cc: opnfv-tech-discuss <[email protected]> Subject: Re: [opnfv-tech-discuss] Hardware Resources in OPNFV +STEINBRUECK, SCOTT A<mailto:[email protected]> Thanks Jack for kicking off this important discussion. I wanted to share my perspective on your questions: >1. What does OPNFV actually plan to release? This one is in a state of flux. Originally, CNTT thought that the output of OPNFV was basically the test suites providing the conformance tests (RC). NOW there is talk in CNTT about how they want the deployment experience to happen - some reference an "EASY" button for deploying a reference implementation. We know that OpenStack has so many configuration alternatives that there is no "one-size-fits-all" pre-configured NFVI. Having said that, there may be value to the VNF vendors in supplying a VNF_test_basis scenario or reference implementation - that is easy to deploy in LaaS and would be a "minimum requirement" for VNF execution. I know even this concept is hard to pin down - I'm just sharing information and brainstorming ideas. [alec] A push button deployment of a RI is clearly desirable and possible with the right amount of work. However, as Jim rightly says, there are multiple types of openstack deployments and each type has its own specific use case: edge, data center + all the different variations in between in terms of size and also in terms of features. It is clearly interesting to be able to run any VNF workload on a RI in a lab, but we need to keep in mind the limitations of such setting. Outside of a RI deployer, OPNFV also releases software tools which may or may not follow closely the OPNFV release schedule. >2. When looking at hardware resources... Big fan of LaaS - but I don't get a vote on this as I'm not a developer :-) [alec] I have very specific requirements regarding LaaS in order to be useful for my project and they are really not that hard to achieve. We can start a separate thread with the LaaS team on this topic (email/slack works best for me as I’m in PST) >3) What does supporting CNTT from a hardware resources perspective look like? CNTT has some vision of using HW for their own needs and are meeting with packet.net<http://packet.net> to discuss this (RI2 related). I do not believe CNTT plans for how to use HW are very mature yet. Scott (added to the distro) is the CNTT TSC chair - perhaps he can be the collection point for CNTT HW needs? [alec] Packet.net is pretty good as we have actually been able to validate my project on packet.net with some collaboration with their infra team. I think if LaaS gets to the same level, that would be awesome, more specifically in the area of APIs to setup the bare metal resources (not only linux server but also extended to the switch). I can provide exactly what APIs we used. My hope is that I can use LaaS as a way to automate the CI for my project as a first step. Until now, validation has been done off OPNFV lab on resources of participating contributors. To Cedric’s point, I don’t really care if the LaaS are reserved statically or on demand, as long as I can run my CI on it with reasonable waiting time, I’m happy. Once LaaS can host any RI, it should not be too hard to automate the data plane benchmarking of that RI using NFVbench like it is being done in a lot of labs outside of OPNFV - this can be integrated in Functest, we’re just missing the LaaS resource requirements/reservation part. Thanks Alec Jim On Wed, Aug 12, 2020 at 4:34 PM Jack Morgan <[email protected]<mailto:[email protected]>> wrote: OPNFV Community, Now that we have made progress on an OPNFV Release Process, I think its time to review our hardware resources strategy within OPNFV. I'm proposing this topic be added to the Weekly Technical Discuss meeting for next Monday, August 17th but wanted to start the discussion via email before hand. OPNFV has had hardware resources for development/testing and releases supporting environments when we had several OPNFV installer projects. Static allocation of hardware resources to an OPNFV project was needed and approaches to have dynamically allocatable hardware resources like LaaS or PDF/SDF were created to help us scale during release time. More recently with just one installer and changes to how/when we release, its time to rethink this approach. Understanding what our overall requirements for hardware resources within the community is needed - not just for development/testing but also supporting releases. Here are a few discussion points to address a couple gray areas. 1. What does OPNFV actually plan to release? 2. When looking at hardware resources... a) What does hardware resources allocation look like for self release model? b) Should OPNFV move to a LaaS model for supporting releases? c) What are our current hardware resources that support releases? 3) What does supporting CNTT from a hardware resources perspective look like? The OPNFV Infra WG (Releng, Pharos and LaaS) is begging to look at this topic. Please share your perspective and input. Regards, -- Jack Morgan -- Jim Baker Linux Foundation Networking - Technical Program Manager mobile: +1 970 227 6007
-=-=-=-=-=-=-=-=-=-=-=- Links: You receive all messages sent to this group. View/Reply Online (#24307): https://lists.opnfv.org/g/opnfv-tech-discuss/message/24307 Mute This Topic: https://lists.opnfv.org/mt/76157120/21656 Group Owner: [email protected] Unsubscribe: https://lists.opnfv.org/g/opnfv-tech-discuss/unsub [[email protected]] -=-=-=-=-=-=-=-=-=-=-=-
