Hi all, I'd like to remind all OPNFV TSC members to provide comments on the Dovetail scope by the end of this week, so that the team can address those before the project comes again to TSC review.
Here are my thoughts. As I see it, and as has been said earlier, the main value of the Certification and Verification Program is to make it easier to run VNF applications on CVP compliant platforms. This includes the use case of testing and benchmarking an NFVi platform in operator premises. Thus, the platform must support - functional requirements like CPU pinning, CPU topology, PCI topology awareness, different overlays as required (VLAN, VXLAN, Geneva,...), different connections to outside (L2 and L3 VPNs), diagnostics, and real-time telemetry - non-functional requirements like networking throughput and latencies, low scheduling latencies, sufficient storage performance for some applications, scalable networking in terms of number of flows, high availability, overall scalability and small minimal configurations We would also like to test behaviors and not the implementation and not the APIs. Looking at my list of functional requirements, many have been already been done in mainstream OpenStack. The different overlays are used by the installers, so it is part of the description of the SUT. The L2 and L3 VPNs are not ready to included, and I have no clear idea about how to test diagnostics and telemetry. As for non-functional requirements, it has already been agreed that anything performance related is not (yet) in Dovetail scope, since it is difficult to decide where to draw the line of acceptable performance. High availability is clearly important, and it is in the scope already. Scalability testing and stress testing are not yet ready for inclusion, but I would put them on the Dovetail roadmap. So it seems that the only thing remaining from my list is networking scalability: launching a number of VMs and setting up a number of flows between them. This is mainly interesting from performance point of view, so it can also be postponed. If I have a platform that passes the proposed Dovetail tests, it will have a recent OpenStack, it has a working HA configuration so that it survives the failure of any OpenStack service, and it is configured so that I can launch VMs and connect to them. I will not know if it has the basic OVS or something accelerated, but otherwise it seems pretty ok to me. I will have to figure out configure external access for my VMs (since the different OPNFV installers behave differently in this) and I may not have familiar systemd commands available, but these are not in OPNFV scope. So, overall, I am pretty ok with Dovetail as it is. Comments? -Tapio _______________________________________________ opnfv-tech-discuss mailing list opnfv-tech-discuss@lists.opnfv.org https://lists.opnfv.org/mailman/listinfo/opnfv-tech-discuss
