This is an old revision of the document!
Day 1 / Track 3 / Talk 3 Workshop: FRRouting Workshop Chair: Donald Sharp and David Lamparter Report by: Donald Sharp
The first presentaion was from Donald Sharp and David Lamparter - cumulus New stuff introduced was nexthop groups (David Ahern was doing this code) and learned later on in zebra session how this is just the start.
A BGP memory rework was discussed where the momery footprint was reduced from 500 MByte to 450 MByte for full internet BGP feed nice. In regards to VRF vs. name-spaces, it would be preferred to get rid of complexity of using name-spaces as VRFs.
A comment received was that name-spaces likely only used earlier on instead of VRF, because VRF wa only implemented much later in Linux. Logically now to use VRF as VRFs and accordingly treat in Linux. Can always run different instances of FRR in different name-spaces.
Presentation 2: session about data-plane / zebra afterwards:
voltanet.io presenter, not the coder himself though. - ZAPI - API between zebra and routing daemons.
Now one thread per client in ZEBRA.
- Aynchronuous communication with FIB. E.g.: remote switch (voltanet).
two threads now: zebra-main, zebra-dplane connected by two queues. zbra dplane talks with "kernel".
- ACK path (installed in HW): only used by sharpd (installs 1 million
routes, testing daemon), but needed for BGP, RIP according to donald. cumulus linux: currently no feedback from linux-kernel when route is installed onto ASIC (eg: broadcom).
- kiran: could expand dplane daemon to directly use chip API to install
routes, not relying on linux-kernel driver doing it. A: Yes.
- toerless: change nexthop-group-members ? Donald: No, not there yet.
Site: https://www.netdevconf.org/0x13/session.html?workshop-FRRouting Slides: Videos: