User Tools

Site Tools


0x13:reports:d1t3t04-an-asynchronous-dataplane-for-frr

Day 1 / Track 3 / Talk 4 Talk – Nuts-n-Bolts: An Asynchronous Dataplane for FRR Speakers: Mark Stapp Report by: Donald Sharp

This session was a follow-up from the previous FRR workshop and discussed the asynchronous data-plane / zebra.

The presenter was from voltanet.io, however it was not the coder himself. The talk begun with the ZAPI - API between zebra and routing daemons. Now it has one thread per client in ZEBRA. This was followed by asynchronous communication with FIB. E.g.: remote switch (voltanet). There are two threads now: zebra-main, zebra-dplane connected by two queues. The zebra dataplane talks with “kernel”. The ACK path (installed in HW) is only used by sharpd (installs 1 million routes, testing daemon), but needed for BGP, RIP according to Donald.

A comment from cumulus linux, there is currently no feedback from linux-kernel when route is installed onto ASIC (eg: broadcom). Kiran asked whether the dataplane deamon could expand to directly use chip API to install routes, not relying on linux-kernel driver doing it. The answer was yes. Toerless then asked to change nexthop-group-members? Donald replied that it's not there yet.

Site: https://www.netdevconf.info/0x13/session.html?talk-FRR-async

0x13/reports/d1t3t04-an-asynchronous-dataplane-for-frr.txt · Last modified: 2019/09/28 17:04 by 127.0.0.1

Donate Powered by PHP Valid HTML5 Valid CSS Driven by DokuWiki