Review plans and diagrams for bare metal deployment of OpenStack

[lixuna]

---

Initial planning tasks:

• OpenStack installation software => Using Kitchen/Chef https://github.com/openstack/openstack-chef (See #47)
• [x] Initial planning of OpenStack cluster needs for Kubecon keynote CNF testing perspective
• [x] Writeup description and needs
• [x] Create diagrams for OpenStack cluster including network connections
• [x] Create slides with diagrams https://docs.google.com/presentation/d/1Z_Zel2GIfZ1of2Y67VTDecfQcIRxfavIpksvgYGs2us/edit#slide=id.g44fd9788f6_0_239

---

• [x] @robertstarmer to review plans and diagrams
• [x] Ask questions as needed
• [x] TBD: Schedule meeting to review plans and needs
• [x] Add additional tasks needed for implementation

[taylor]

Cluster information and Q/A

How many nodes OpenStack cluster? What their high-level purposes? What other systems?

• 5 nodes total
• 2 controller nodes
• 3 compute nodes
  • Storage on compute nodes themselves

[FYI ONLY:

• 1 packet generator system on machine w/o OpenStack]

Hardware requirements

• Dataplane CNF Compute nodes (currently all 3) on system with Intel x710 4-port NICs
• Reserved system in Packet.net
• Packet.net machine type m2.xlarge.x86
• Intel 5120 Gold CPU

Network configuration needs/request

Functionality:

• Required: External/public network access from all VMs (eg. downloads from internet work)
• Required: Host systems can access all VMs via SSH
• Required: Public internet SSH access to compute and control host systems
• Required: VMs on same host connect to each other over vhost through VPP vSwitch
• Nice to have: Inbound Public Internet connection to all VMs (eg. SSH)

Packet.net networks (created outside of OpenStack)

• Public internet (default Packet.net L3) connected to all openstack nodes and packet generator
• One Packet.net layer-2 (physical) networks connected to all systems used as OpenStack management network
• Two Packet.net layer-2 (physical) networks connected to all OpenStack machines and Packet generator system

OpenStack networks for Virtual Network Function dataplane traffic

• Two tenant networks connected to Packet.net layer-2 networks
• VPP vSwitch on each compute node bridging the two Packet.net layer-2 networks
• VMs connected to VPP vSwitch over vhost

What network options do we have for the dataplane NFs

• Preferred: VPP Neutron plugin (eg. https://github.com/openstack/networking-vpp)
  • https://vtalks.net/talk/networking-vpp-a-fast-forwarding-vswitchvrouter-for-openstack/
  • http://superuser.openstack.org/articles/vpp-vswitchvrouter-openstack/
• Risky/Unknown: VPP manually configured to work with OpenStack VMs/neutron
• (NOT ACCEPTABLE) VMs connected to physical L2 network via public bridge on tap interfaces
  • https://wiki.onap.org/display/DW/ONAP+vCPE+VPP-based+VNF+Installation+and+Usage+Information
  • Not good for performance
• SR-IOV to VMs (Virtual Functions mapped directly to VMs)
  • https://docs.openstack.org/mitaka/networking-guide/config-sriov.html,
  • https://docs.openstack.org/ocata/networking-guide/config-sriov.html,
  • https://wiki.openstack.org/wiki/SR-IOV-Passthrough-For-Networking
  • Limited on number of Network Virtual Functions the network card supported
  • Chaining VNFs network connections will have limited configuration (eg. basic switching at NICs)

Deferred configuration needs

• CPU pinning
• NUMA configuration

[taylor]

5 node OpenStack Cluster + packet generator

• 6 Total Packet.net machines
• All connected to the two layer 2 Packet.net networks
• Number of worker machines should be configurable at OpenStack deployment time
  • Scale from 1 worker node up to how ever many are specified

---

Physical NIC connected to Packet.net and host connections

• Three layer-2 networks will be setup in Packet.net’s infrastructure
• 2 layer-2 networks will be controlled and configured by VPP vSwitch
• One layer-2 network will be configured by OpenStack in the Linux kernel for the management network
• One connection will provide public internet access and be configured by the Linux kernel networking

---

VPP vSwitch and OS ports

• VPP will be attached to two tenant networks for the compute node
• VPP will be attached to a bridged tenant network connecting each set of VMs

[taylor]

@robertstarmer ready for your review.

[robertstarmer]

@lixuna @taylor Reviewing.

[lixuna]

@robertstarmer checking in, please let us know if you would like to schedule a call to answers any questions you may have.

[lixuna]

Follow-up call scheduled for Tuesday, October 23.

[lixuna]

@robertstarmer please ping me and @taylor if we can review any documentation prior to the follow-up call.