eBPF-based per-packet multipath routing.
Q: How does routing work?
erezd is a daemon that runs on metals in encap mode, and on edge
routers in decap mode.
- On metals, the daemon peers with BGP routers to collect all nexthops for each prefix. On every outbound packet it picks a nexthop at random and encapsulates the packet in IPv6/GRE toward it
- On edge routers, the daemon receives these encapsulated packets, strips the IPv6/GRE header, and forwards the inner packet through the nexthop
Currently nexthop selection is random (!), but support is planned for balancing based on capacity and health, in addition to other metrics.
Q: Is this production-ready?
Not yet!