Note
This application is distinct from the use of BaGPipe to create IPVPN or E-VPN interconnections in the context of the BGPVPN Interconnection API (see below).
The bagpipe
mechanism driver for Neutron’s ML2 core plugin, when enabled
along with the corresponding compute node agent extension, will result in
Neutron tenant networks to be realized with E-VPN/VXLAN.
How it works is that a BGP VPN identifier (called a BGP “Route
Target”) will be defined for each Neutron tenant network, derived from the
VXLAN VNI a.k.a segmentation ID, and the bagpipe
agent extension on compute
nodes will setup a corresponding E-VPN instance with
this identifier on the local BaGPipe-BGP instance on the compute node and
attach VM ports to this instance as needed.
This solution is currently supported with the linux networking stack (i.e. with
the linuxbridge
agent enabled with bagpipe extension, and BaGPipe-BGP driver
for the linux bridge VXLAN implementation). The approach would be easily
extended to support OpenVSwitch as well.
Another way to understand this approach for someone coming with a Neutron ML2 background is that it is similar to the l2population mechanism except that the bridge forwarding entries are populated based on BGP VPN routes rather than based on information distributed in RPCs. This similarity comes with a difference: while l2population announces the information on one messaging topic, each compute node receiving information about all Neutron networks even the ones not present on its vswitch, the behavior with BaGPipe ML2 is that a compute node will only receive the mappings that it needs.
Note
This application is distinct from the use of BaGPipe to realize
Neutron networks with BGP E-VPNs. bagpipe
driver for
networking-bgpvpn supports both IPVPNs and E-VPNs, but does not rely on
bagpipe
ML2 mechanism driver to do so.
In this application, networking-bagpipe
aims at proposing a lightweight
implementation of the BGPVPN Interconnection service, designed to work with
the ML2 openvswitch
or linuxbridge
mechanism drivers (or as an
alternative with the bagpipe
ML2 mechanism driver).
When used along with the openvswitch
or linuxbridge
ML2 mechanism
driver, it involves the use of:
bagpipe
driver for the BGPVPN service plugin (in networking-bgpvpn package)bagpipe_bgpvpn
extension for the Neutron compute node agent
(in this package)Example with OVS agent:
For this application, networking-bagpipe
provides a bagpipe
driver
for the networking-sfc
that will result in service chains defined via the
networking-sfc API, being realized with BGP VPN stiching, BGP VPN route
redistribution and BGP Flowspec routes.
The components involved are:
bagpipe
driver for the networking-sfc
service plugin (in this package)bagpipe_sfc
extension for the Neutron compute node agent
(in this package)Note
This driver is still quite experimental, and still currently
relies on using the linuxbrige
agent along with the OVS dataplane
driver for IPVPN in bagpipe-bgp
.
Work in progress:
Considered:
Except where otherwise noted, this document is licensed under Creative Commons Attribution 3.0 License. See all OpenStack Legal Documents.