Why Is BFD More Light Weight Than Routing Hellos?

There are many articles on BFD. It is well known that BFD has the following advantages over routing protocol hellos/keepalives:

  • BFD is more light weight than hellos/keepalives.
  • Multiple clients can register to BFD instead of configuring each protocol with aggressive timers.
  • On some platforms, BFD can be offloaded to the hardware instead of the CPU.
  • BFD provides faster timers than routing protocols.
  • BFD is less CPU intensive.

What does light weight mean, though? Does it mean that the packets are smaller? Let’s compare a BFD packet to an OSPF Hello. Starting with the OSPF Hello:

Frame 269: 114 bytes on wire (912 bits), 114 bytes captured (912 bits) on interface ens192, id 1
Ethernet II, Src: 00:50:56:ad:8d:3c, Dst: 01:00:5e:00:00:05
Internet Protocol Version 4, Src:, Dst:
Open Shortest Path First
    OSPF Header
        Version: 2
        Message Type: Hello Packet (1)
        Packet Length: 48
        Source OSPF Router:
        Area ID: (Backbone)
        Checksum: 0x7193 [correct]
        Auth Type: Null (0)
        Auth Data (none): 0000000000000000
    OSPF Hello Packet
    OSPF LLS Data Block

There’s 114 bytes on the wire consisting of:

Applying BGP Policy Templates

I got this question after publishing the BGP Session Templates lab exercise:

Would you apply BGP route maps with a peer/policy template or directly to a BGP neighbor? Of course, it depends; however, I believe in using a template for neighbors with the same general parameters and being more specific per neighbor when it comes to route manipulation.

As my reader already pointed out, the correct answer is It Depends, now let’s dig into the details ;)

Network Layer: Interface or Node Addresses

The fun question about network layer addresses is: are we addressing nodes or individual node interfaces? On the data link layer, we never had this issue because it was obvious that a data link layer endpoint is an interface, so each interface should have a unique data link layer address.

Interestingly, that’s not the case on transparent bridges. Even though they have multiple interfaces, the whole bridge has a single MAC address, so one could claim we’re addressing nodes connected to a single data link layer. The IEEE standard is unambiguous: in every relevant diagram, the MAC address sits on top of multiple interfaces because the MAC address belongs to the control plane.

All you need to know about the Digital Services Act

February 17th, 2024 marked the entry into force of a landmark piece of European Union (EU) legislation, affecting European users who create and disseminate online content as well as tech companies who act as “intermediaries” on the Internet. I am talking of course about the EU Digital Services Act, or DSA for short. The DSA was first proposed in December 2020, and is meant to update a 20-year-old law called the EU e-commerce Directive, which provides important safeguards and legal certainty for all businesses operating online. The principles of that legal framework, most notably the introduction of EU-wide rules on intermediary liability, are still of major importance today. The DSA is a landmark piece of European legislation because it also sets out, for the first time, enhanced regulatory requirements for (large) digital platforms, thus affecting the entire Internet ecosystem.

At Cloudflare, we are supportive of the longstanding legal frameworks both in Europe and other parts of the world that protect Internet companies from liability for the content that is uploaded or sent through their networks by their users, subscribers or customers. These frameworks are indispensable for the growth of online services, and have been essential in the growth Continue reading

Catalyst SD-WAN Enhanced Application Aware Routing

Traditionally, Cisco has leveraged BFD to monitor tunnels and their performance and Application Aware Routing (AAR) to reroute traffic. BFD has been used to measure:

  • Latency.
  • Loss.
  • Jitter.

Additionally, BFD is also used to verify liveliness of the tunnels. This works well, but there are some drawbacks to using a separate protocol for measuring performance:

  • You are adding control plane packets competing for bandwidth with packets in data plane.
  • Sending control plane packets frequently may overload the control plane.
    • This may lead to false positives.
  • It’s not guaranteed that control plane packets and data plane packets are treated equally.
  • AAR did take some time to react to poor transports as it had to collect enough measurements before reacting.
  • AAR didn’t have a built-in dampening mechanism.

With the default BFD settings, BFD packets are sent every second. The default AAR configuration consists of six buckets that hold 10 minutes of data each. This means that with the default settings, AAR will react in 10-60 minutes depending on how poorly the transport is performing. The most aggressive AAR configuration recommended by Cisco was to have 5 buckets holding 2 minutes of data each. AAR would then react in 2-10 minutes which I Continue reading

Boat build phase – Part1

Now I had a design and a half empty shell it was time to get work. The biggest problem with working on a boat is space, you are forever moving things and constantly tidying up. All the materials had to be orderd to arrive just in time for when they were needed as there is no space on the boat to store them. The remoteness of the marina and not having a car didn’t help, I wasted a lot of hours getting local buses to go hire a van or waiting on deliveries.

VPP on FreeBSD – Part 2

About this series


Ever since I first saw VPP - the Vector Packet Processor - I have been deeply impressed with its performance and versatility. For those of us who have used Cisco IOS/XR devices, like the classic ASR (aggregation services router), VPP will look and feel quite familiar as many of the approaches are shared between the two. Over the years, folks have asked me regularly “What about BSD?” and to my surprise, late last year I read an announcement from the FreeBSD Foundation [ref] as they looked back over 2023 and forward to 2024:

Porting the Vector Packet Processor to FreeBSD

Vector Packet Processing (VPP) is an open-source, high-performance user space networking stack that provides fast packet processing suitable for software-defined networking and network function virtualization applications. VPP aims to optimize packet processing through vectorized operations and parallelism, making it well-suited for high-speed networking applications. In November of this year, the Foundation began a contract with Tom Jones, a FreeBSD developer specializing in network performance, to port VPP to FreeBSD. Under the contract, Tom will also allocate time for other tasks such as testing FreeBSD on common virtualization platforms to improve the desktop experience, improving Continue reading

Keeping the Pipe Just Full

In this post, I will be discussing a paper published by the Internet pioneer Leonard Kleinrock, titled “Keep the Pipe Just Full, But No Fuller”. The paper’s conclusion is that it is best to keep the internet “pipe” full, without overloading it. This idea is a take on Einstein’s famous quote, “Make everything as simple as possible, but not simpler.”


It’s not always true that using the network to its fullest capacity will lead to better performance. When the network is overloaded, congestion and queueing delays can occur, which can affect the rate at which useful data is delivered and the time it takes for data to be delivered. This means there is a tradeoff between goodput and latency. To avoid network issues, congestion management is important. TCP regulates data flow from source to destination by sending packets while monitoring the delivery rate and response time to manage congestion. The challenge is to regulate traffic flow, as both underutilizing and overloading the capacity can waste resources and cause congestion.

For many years, TCP congestion control algorithms have relied on loss as a measure of congestion. However, loss causes sawtooth behaviors as flow expands until packets are dropped, then Continue reading

Hedge 213: Batfish with Ratul Mahajan

Network configuration analysis has always been the domain of commercial-grade software. Batfish changes all that with an open source, community-supported tool that can find errors and guarantees the correctness of planned or current network configurations. Ratul Mahajan joins Tom Ammon and Russ White to talk about this new tool, its capabilities, and the importance of network configuration analysis.

You can find out more about Batfish at the project home page, and in this paper by Ratul.

Monitoring machine learning models for bot detection

Cloudflare’s Bot Management is used by organizations around the world to proactively detect and mitigate automated bot traffic. To do this, Cloudflare leverages machine learning models that help predict whether a particular HTTP request is coming from a bot or not, and further distinguishes between benign and malicious bots. Cloudflare serves over 55 million HTTP requests per second — so our machine learning models need to run at Cloudflare scale.

We are constantly making improvements to the models that power Bot Management to ensure they are incorporating the latest threat intelligence. This process of iteration is an important part of ensuring our customers stay a step ahead of malicious actors, and it requires a rigorous process for experimentation, deployment, and ongoing observation.

We recently shared an introduction to Cloudflare’s approach to MLOps, which provides a holistic overview of model training and deployment processes at Cloudflare. In this post, we will dig deeper into monitoring, and how we continuously evaluate the models that power Bot Management.

Why monitoring matters

Before bot detection models are released, we undergo an extensive model testing/validation process to ensure our detections perform as expected. Model performance is validated across a wide number of web traffic Continue reading