Repost: L2 Is Bad

Roman Pomazanov documented his thoughts on the beauties of large layer-2 domains in a LinkedIn article and allowed me to repost it on ipSpace.net blog to ensure it doesn’t disappear

First of all: “L2 is a single failure domain”, a problem at one point can easily spread to the entire datacenter.

Repost: L2 Is Bad

Roman Pomazanov documented his thoughts on the beauties of large layer-2 domains in a LinkedIn article and allowed me to repost it on ipSpace.net blog to ensure it doesn’t disappear

First of all: “L2 is a single failure domain”, a problem at one point can easily spread to the entire datacenter.

BGP Labs: Protect EBGP Sessions

I published another BGP labs exercise a few days ago. You can use it to practice EBGP session protection, including Generalized TTL Security Mechanism (GTSM) and TCP MD5 checksums¹.

I would strongly recommend to run BGP labs with netlab, but if you like extra work, feel free to use any system you like including physical hardware.

BGP Labs: Protect EBGP Sessions

I published another BGP labs exercise a few days ago. You can use it to practice EBGP session protection, including Generalized TTL Security Mechanism (GTSM) and TCP MD5 checksums¹.

Addresses in a Networking Stack

After discussing names, addresses and routes, it’s time for the next question: what kinds of addresses do we need to make things work?

End-users (clients) are usually interested in a single thing: they want to reach the service they want to use. They don’t care about nodes, links, or anything else.

End-users might want to use friendly service names, but we already know we need addresses to make things work. We need application level service identifiers – something that identifies the services that the clients want to reach.

Addresses in a Networking Stack

After discussing names, addresses and routes, it’s time for the next question: what kinds of addresses do we need to make things work?

End-users (clients) are usually interested in a single thing: they want to reach the service they want to use. They don’t care about nodes, links, or anything else.

End-users might want to use friendly service names, but we already know we need addresses to make things work. We need application level service identifiers – something that identifies the services that the clients want to reach.

Names, Addresses and Routes

It always helps to figure out the challenges of a problem you’re planning to solve, and to have a well-defined terminology. This blog post will mention a few challenges we might encounter while addressing various layers of the networking stack, from data-link layer and all the way up to the application layer, and introduce the concepts of names, addresses and routes.

According to Martin Fowler, one of the best quotes I found on the topic originally came from Phil Karlton:

Names, Addresses and Routes

It always helps to figure out the challenges of a problem you’re planning to solve, and to have a well-defined terminology. This blog post will mention a few challenges we might encounter while addressing various layers of the networking stack, from data-link layer and all the way up to the application layer, and introduce the concepts of names, addresses and routes.

According to Martin Fowler, one of the best quotes I found on the topic originally came from Phil Karlton:

Dataplane MAC Learning with EVPN

Johannes Resch submitted the following comment to the Is Dynamic MAC Learning Better Than EVPN? blog post:

I’ve also recently noticed some vendors claiming that dataplane MAC learning is so much better because it reduces the number of BGP updates in large scale SP EVPN deployments. Apparently, some of them are working on IETF drafts to bring dataplane MAC learning “back” to EVPN. Not sure if this is really a relevant point - we know that BGP scales nicely, and its relatively easy to deploy virtualized RR with sufficient VPU resources.

While he’s absolutely correct that BGP scales nicely, the questions to ask is “what is the optimal way to deliver a Carrier Ethernet service?”

Dataplane MAC Learning with EVPN

Johannes Resch submitted the following comment to the Is Dynamic MAC Learning Better Than EVPN? blog post:

I’ve also recently noticed some vendors claiming that dataplane MAC learning is so much better because it reduces the number of BGP updates in large scale SP EVPN deployments. Apparently, some of them are working on IETF drafts to bring dataplane MAC learning “back” to EVPN. Not sure if this is really a relevant point - we know that BGP scales nicely, and its relatively easy to deploy virtualized RR with sufficient VPU resources.

While he’s absolutely correct that BGP scales nicely, the questions to ask is “what is the optimal way to deliver a Carrier Ethernet service?”

Worth Reading: Where Are the Self-Driving Cars?

Gary Marcus wrote an interesting essay describing the failure of self-driving cars to face the unknown unknowns. The following gem from his conclusions applies to AI in general:

In a different world, less driven by money, and more by a desire to build AI that we could trust, we might pause and ask a very specific question: have we discovered the right technology to address edge cases that pervade our messy really world? And if we haven’t, shouldn’t we stop hammering a square peg into a round hole, and shift our focus towards developing new methodologies for coping with the endless array of edge cases?

Obviously that’s not going to happen, we’ll keep throwing more GPU power at the problem trying to solve it by brute force.

Worth Reading: Where Are the Self-Driving Cars?

Gary Marcus wrote an interesting essay describing the failure of self-driving cars to face the unknown unknowns. The following gem from his conclusions applies to AI in general:

In a different world, less driven by money, and more by a desire to build AI that we could trust, we might pause and ask a very specific question: have we discovered the right technology to address edge cases that pervade our messy really world? And if we haven’t, shouldn’t we stop hammering a square peg into a round hole, and shift our focus towards developing new methodologies for coping with the endless array of edge cases?

Obviously that’s not going to happen, we’ll keep throwing more GPU power at the problem trying to solve it by brute force.

BGP Labs: Bidirectional Route Redistribution

In the next BGP labs exercise you’ll build the customer part of an MPLS/VPN solution. You’ll use bidirectional OSPF-to-BGP route redistribution to connect two sites running OSPF over a Service Provider MPLS backbone.

I would strongly recommend to run labs with netlab, but if you like extra work, feel free to use any system you like including physical hardware.

BGP Labs: Bidirectional Route Redistribution

In the next BGP labs exercise, you’ll build the customer part of an MPLS/VPN solution. You’ll use bidirectional OSPF-to-BGP route redistribution to connect two sites running OSPF over a Service Provider MPLS backbone.

DHCP Relaying Across a Firewall

Chinar Trivedi wanted to know what happens when you insert a firewall in the DHCP data path (original question.

TL&DR: Nothing much, but that does not mean you should.

Now for the details:

DHCP Relaying Across a Firewall

Chinar Trivedi wanted to know what happens when you insert a firewall in the DHCP data path (original question.

TL&DR: Nothing much, but that does not mean you should.

Now for the details:

OSPF ECMP with Unnumbered IPv4 Interfaces

The OSPF and ARP on Unnumbered IPv4 Interfaces triggered an interesting consideration: does ECMP with across parallel unnumbered links?

TL&DR: Yes, it works flawlessly on Arista EOS and Cisco IOS/XE. Feel free to test it out on any other device on which netlab supports unnumbered interfaces with OSPF.

OSPF ECMP with Unnumbered IPv4 Interfaces

The OSPF and ARP on Unnumbered IPv4 Interfaces triggered an interesting consideration: does ECMP work across parallel unnumbered links?

TL&DR: Yes, it works flawlessly on Arista EOS and Cisco IOS/XE. Feel free to test it out on any other device on which netlab supports unnumbered interfaces with OSPF.

Reliable ECMP with Static Routing

One of my readers wanted to use EIBGP (hint: wrong tool for this particular job¹) to load balance outgoing traffic from a pair of WAN edge routers. He’s using a design very similar to this one with VRRP running between WAN edge routers, and the adjacent firewall cluster using a default route to the VRRP IP address.

The problem: all output traffic goes to the VRRP IP address which is active on one of the switches, and only a single uplink is used for the outgoing traffic.

Reliable ECMP with Static Routing

One of my readers wanted to use EIBGP to load balance outgoing traffic from a pair of WAN edge routers (hint: wrong tool for this particular job¹). He’s using a design very similar to this one with VRRP running between WAN edge routers, and the adjacent firewall cluster using a default route to the VRRP IP address.

The problem: all output traffic goes to the VRRP IP address which is active on one of the switches, and only a single uplink is used for the outgoing traffic.