Cloud Native Computing Foundation KubeCon + CloudNativeCon Europe 2017 (Berlin), 6 Apr 2017

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Lightning Talk: BGP Routing in Kubernetes - Pankaj Sehgal, Tata Consultancy Services

Description

Lightning Talk: BGP Routing in Kubernetes - Pankaj Sehgal, Tata Consultancy Services

Kubernetes imposes some requirements on networking of pods/containers. You cannot take two machines running containers and expect Kubernetes to work. There are number of ways we may make it work, but there is much simpler custom BGP Protocol develop to achieve this.

About Pankaj Sehgal
I have been working with Apple as a consultant for past 5 years. Major focus is always on round the clock availability of icloud services. I love working with new micro technologies and running in my free time.
Join us for KubeCon + CloudNativeCon in Barcelona May 20 - 23, Shanghai June 24 - 26, and San Diego November 18 - 21! Learn more at https://kubecon.io. The conference features presentations from developers and end users of Kubernetes, Prometheus, Envoy and all of the other CNCF-hosted projects.

A

Please welcome to the stage Pankaj deck hall of the Tata Consultancy Services.

A

Hi folks, so my name is pankaj and I'm gonna talk about be cheap in kubernetes, so first jumping into the bgp part of the Kuban it is. We have been deploying containers in production and I'm sure like we are running thousands of containers. So the orchestration part is almost taken care by cuban artists, but there is still a gray area which there's still a grey area for the networking, which is still we don't have any standardized thing so that working there are protocols. There is everything that people have been deploying.

A

So, first of all, why bgp and I'll take a step back I'll. First of all, I will tell come on yeah yeah. Why not other protocols? So let's start here, this is my ideal Network. You have a router, you have a number of endpoints and say if I want to send an IP packet from end point, one to end point two: it will go through a router. So what basically router doing here is its policing, my IP networks, so it's policing my IP packets and that's it.

A

So if the network is having X number of endpoints- and you have just one router you're fine now when I have two autonomous systems, if I want to send IP packet from autonomous system from one to autonomous system from two, it will go via the routers. What now happening is my route. My routers are having more informations and, as my endpoints are increasing, the information in the router is also increasing and I jammed. It so think of it like this.

A

If you have just a couple of endpoints in your network, your router will be having that small information. But if the number of endpoints in the network are increasing rapidly, you are scaling out Aspen in the production. Your routers will be having more data the more time to process your IP packet I know it's a microsecond or a millisecond, but when it counts to the traffic, when you say the amount of traffic it actually counts, so BGP solves their problem.

A

So what bgp does is BGP has nothing. It has an IP table, it has entries for the direct hopes and it has if the hope is more than if there is move on more hope, then it has the hope for a next BGP pair, and that's it that's what BGP does. So, if you want to send your IP packet.

A

That's it so this is my an ideal Cuban, it's deployment, you have a Cuban each master. You have a couple of nodes in which the IP tables are there I'm using a Linux kernel to do my routing.

A

Let's say one of my system is crashed. Now what will happen? Is the routing tables will get updated when I say updation I'm, actually deleting the entry of the system that actually crashed? So if I have ten endpoints I'm actually doing ten deletion operations but notice one thing over here: how I am Bill eating those the peers are talking to each other? Peers are talking to each other, and one pair is telling so one by endpoint is telling another end point. Okay, this is down delete the entry from the IP table.

A

This is down delete the IP from the entry table, and this is fine again if your endpoints and 2050- but we are talking about thousands of nodes here which are running highly tightly coupled containers over them and something like this will happen through the routing tables. You don't want to do that.

A

So very simple and stupid solution, for this could be when any of my system crashes, why not feed that info to a route feeder, given its master, will take care of it? Okay, this this particular node, is done. It will update the route feeder. My route feeder will update the entries in the nodes, so what I'm doing here is route feeder after X minute of time it will update the routing tables or it will update the IP tables in all of the nodes.

A

So you don't have any overhead of updating the IP tables in the Linux terminal and, moreover, you can also deploy your security schemes or any other third party that you want to do so I'm trying to make my networking and my routing equal. That's all the only part that now in the networking will be routing, and that's it that's a little though, and you drill update the routing tables. That's all so I encourage you to come and talk to me regarding this.

A

This is just a concept which is being deployed and I'm way too much on slag. You can let's talk about it in another two days, and hopefully it will be a good one. Thank you.