Summary

Distributed virtual routers have a positive impact on the network architecture as a whole by pushing east-west traffic between instances and north-south traffic through floating IPs to the compute nodes, removing bottlenecks and single points of failure seen in the legacy model. Traffic from virtual machines without floating IPs must still traverse a centralized network node when routing to external networks. This is seen as a compromise, however, given that a high number of IPv4 addresses would be required if SNAT were handled at the compute node layer.

While distributed virtual routers help provide parity with nova-network's multi-host capabilities, they are not without their limitations. Work is under way to add support to distributed virtual routers for IPv6, advanced services such as LBaaS, FWaaS, and VPNaaS, conversions from legacy to distributed routers, and more. Implementing a distributed virtual router is transparent to the user, but it is operationally complex and considerably...