Alec, thanks for good inputs from OpenStack use cases perspective! I agree with most if not all points. Suggest we walk thru these on the TWS call, hopefully this Friday.
-Maciek On 29 Nov 2016, at 20:36, Alec Hothan (ahothan) <ahot...@cisco.com<mailto:ahot...@cisco.com>> wrote: Thanks for putting this together! A few preliminary general comments which we can discuss Wednesday. 2p1nic vs 1p1nic: I know lots of vswitch benchmarks love pairing phys interfaces as it is easier to scale with more cores… But a lot of openstack deployments will come with 1 physical port for all the tenant traffic. In that case, looping traffic will come in through a vlan and go out to another vlan (vlan case) or come in through a vxlan tunnel and go out another vxlan tunnel (VxLAN overlay). The maximum throughput is 10G in and 10 G out, less wires on the compute nodes and less ports on the TOR. I’m not sure there are/will be many deployments with 2 independent physical ports for tenant traffic. At 30 nodes per rack that makes it 60 ports per TOR just for the tenant traffic…. If you factor in bonding, that would make it 4 wires per compute node instead of 2. For those deployments that really need more than a 10G link, they might just use 1 40G link (or 2 bonded links) rather than 2 individual 10G links. Neutron supports multiple interfaces but it looks harder to manage (I guess you need to split your vlan range in 2 but in the case of vxlan overlay I’m not sure how that can be managed. I’d like to hear from others whether we should focus on 1p1nic first. It just seems to me this might be more representative of real NFV deployments than 2p1nic. Test topologies of interest for OpenStack: Could we mark all those test topologies that are applicable to OpenStack? Or more generally describe the use case for each test topology. I don’t think OpenStack will use any l2xc in vpp. OpenStack ML2/VPP/VLAN will use dot1q-l2bdbase-eth-2vhost-1vm OpenStack VPP/VxLAN will use ethip4vxlan-l2bdbase-eth-2vhost-1vm VM images: I think vsperf is using testpmd. I know some other perf teams use l2fwd. In our team, we use VPP l2 x-connect in the VM (what you call vswitch VPP) and evaluating testpmd and they don’t seem to differ to much in results. Seems like testpmd is the way to go for L2… Would also be good to have the VM image creation scripted – so that anybody can recreate them from scratch. We use DIB (openstack disk image builder) for creating our VM images but any other scripted solution should work. For easier reuse inside openstack, the config of the VM instance must be as automatic as possible. That is easy for the L2 case (just cross connect between the 2 virtual interfaces). For the L3 case, the L3 config should be done through config drive (not SSH). That’s it for now, Alec From: "Maciek Konstantynowicz (mkonstan)" <mkons...@cisco.com<mailto:mkons...@cisco.com>> Date: Tuesday, November 29, 2016 at 6:27 PM To: Thomas F Herbert <therb...@redhat.com<mailto:therb...@redhat.com>>, vpp-dev <vpp-dev@lists.fd.io<mailto:vpp-dev@lists.fd.io>>, "csit-...@lists.fd.io<mailto:csit-...@lists.fd.io>" <csit-...@lists.fd.io<mailto:csit-...@lists.fd.io>>, "Pierre Pfister (ppfister)" <ppfis...@cisco.com<mailto:ppfis...@cisco.com>>, Andrew Theurer <atheu...@redhat.com<mailto:atheu...@redhat.com>>, Douglas Shakshober <dsh...@redhat.com<mailto:dsh...@redhat.com>>, Rashid Khan <rk...@redhat.com<mailto:rk...@redhat.com>>, Karl Rister <kris...@redhat.com<mailto:kris...@redhat.com>>, Irene Liew <irene.l...@intel.com<mailto:irene.l...@intel.com>>, "Alec Hothan (ahothan)" <ahot...@cisco.com<mailto:ahot...@cisco.com>>, Damon Wang <damon.dev...@gmail.com<mailto:damon.dev...@gmail.com>> Subject: Re: [vpp-dev] vHost user test scenarios for CSIT All, Here is the first draft: https://wiki.fd.io/view/CSIT/vhostuser_test_scenarios I did my best to capture all inputs as per this thread. But it’s hardly readable yet - requires more TLC :) See what you think - feel free to add/edit things directly on FD.io<http://fd.io/> wiki page. Suggest to discuss next steps on csit weekly call tomorrow, details here: https://wiki.fd.io/view/CSIT/Meeting -Maciek On 28 Nov 2016, at 07:37, Thomas F Herbert <therb...@redhat.com<mailto:therb...@redhat.com>> wrote: All, At last week's CSIT meeting, Maciek (mkons...@cisco.com<mailto:mkons...@cisco.com>) offered to compile a summary suggestions on this mailing list. On 11/22/2016 11:34 AM, Pierre Pfister (ppfister) wrote: Hello Thomas, Sorry I haven't reached out faster, I was travelling. Please have a look at vppsb/vhost-test It includes a standalone script which provides VPP and VM configuration for PVP tests. - Runs testpmd in the VM - Supports various CPU configuration for VPP - Can run with or without gdb, debug or release Not committed yet: - Supports VM restart - Support for VPP restart - Support for multiple additional (dead) vhost interface I did that outside of the context of CSIT so people can: - Look at it and see what are the optimisations that are used - Use it without CSIT I will keep using and improving it because I use it for my own development and testing purposes. Rest of this inline. Le 8 nov. 2016 à 22:25, Thomas F Herbert <therb...@redhat.com<mailto:therb...@redhat.com>> a écrit : All: Soliciting opinions from people as to vhost-user testing scenarios and guest modes infd.io<http://fd.io/> CSIT testing of VPP - vhost-user. I will forward to this mailing list as well as summarize any additional feedback. I asked some people that happen to be here at OVSCON as well as some Red Hat and Intel people. I am also including some people that are involved in upstream vhost-user work in DPDK. So far, I have the following feedback with an attempt to condense feedback and to keep the list small. If I left out anything, let me know. In addition to the PVP tests done now with small packets. Testpmd in guest is OK for now. 1 Add multiple VMs (How many?) Makes sense to me. 2 is enough (4 would be good number). 2 Both multi-queue and single-queue Yes. Ideally, 1-2-4 queues. With different number of workers (0 workers, i.e. single VPP thread, 1 worker, queues*2 workers). 3 Tests that cause the equivalent of multiple flows in OVS. Varying variety of traffic including layer 2 and layer 3 traffic. Yes. Should test with L2 and L3. 4 Multiple IF's (Guest or Host or Both?) Possibly. But more importantly, I think, we need to have VM restart and interface restart (delete - create). OpenStack integration generates a significant amount of delete-recreate of vhost interface. The following might not be doable by 17.01 and if not consider the following as a wish list for future: 1 vxLan tunneled traffic 2 VPP in guest with layer 2 and layer 3 vRouted traffic. 3 Additional Overlay/Underlay: MPLS --TFH -- Thomas F Herbert SDN Group Office of Technology Red Hat _______________________________________________ vpp-dev mailing list vpp-dev@lists.fd.io<mailto:vpp-dev@lists.fd.io> https://lists.fd.io/mailman/listinfo/vpp-dev -- Thomas F Herbert SDN Group Office of Technology Red Hat
_______________________________________________ vpp-dev mailing list vpp-dev@lists.fd.io https://lists.fd.io/mailman/listinfo/vpp-dev
