Lessons learnt from Netdev 1.2 in Tokyo

Posted by Zevenet | 10 October, 2016 | Reports, Technical

In this post we’re going to highlight the most important topics and discussions addressed in Netdev 1.2 in Tokyo the last 5th-7th of October to create the networking evolution in the Linux Kernel.

One interesting approach has been the idea of reducing NICs buffers that could provide a great advantage to avoid losses of packets and therefore reduce packets retransmissions saving bandwidth resources. The Google team presented this technique using BBR (Bottleneck bandwidth and round-trip propagation time) + RACK. It’ll be available from linux kernel 4.9.

making_tcp_fast

In regards to the kernel securization support, we can get a great advantage of the kTLS socket to provide channel securization and combined with KCM (Kernel Connection Multiplexor) sockets we could get the cpus cycles reduced by 93/100.

Verizon stated in their keynote that they’ve been moving their data centers infrastructure to Linux thanks to the flexibility and accessibility solutions after using proprietary software for years.

verizon

eXpress Data Path has been presented with great expectation and Dave Miller stated that “DPDK is not Linux. XDP is.”, which it’s a hook that the driver needs to support and then execute a packet action even before the skb kernel structure is created.

xdp

Tom Herbert presented their implementation on Facebook infrastructure of Datacenter Networking Stack (DCTCP) where IPv6 is a must, encryption/decryption on the fly is a goal and stream parsing a reality.

herbert

Cumulus Networks presented some of their great work with the ifupdown2 command for Debian systems which provides advanced networking functions for bridging and switching, and VRF (Virtual Routing & Forwarding) project to isolate containers communications.

Mellanox provided HW offload solutions for almost everything!

Sophos presented an interesting approach using a fast path packet marks in order to improve forwarding bandwidth with great results.

Finally, the nftables workshop presented the most important changes and improvements since the last Netdev, where Pablo Neira shown the way to play lego with nftables and it’s 22 instructions.

In this nftables workshop, Laura García from Zen Load Balancer Team presented the how create a load balancer 10 times faster in Direct Server Return with nftables from ingress than LVS.

Now it’s the time to say bye bye Tokyo, but see you soon until the next Netdev 2.1 in Montreal, Canada!

CHECK OUR NETDEV TALKS PAPERS

SHARE ON:

Related Blogs

Posted by zenweb | 07 June 2018
User experience doesn't have to be a limitation to access to high technology. For that reason at Zevenet we take care usability and management easiness to provide cutting edge networking…
Posted by zenweb | 30 May 2018
Load Balancers are a key device in our infrastructure as they are used to distribute traffic across different networks, they are the entry point of applications and even of a…
Posted by zenweb | 23 May 2018
The transition to IPv6 is inevitably close, organizations around the world are preparing their systems, operations and procedures in order to be ready when the time comes. Some of them…