Introduction

The Advanced Networking course focuses on recent development of communication networks and Internet protocols. We take a hands-on approach, with many of the assignments involving practical experimentation and network programming tasks. The course is divided in to 9 parts, including this introduction. Each part comes with an assignment.

The hands-on assignments assume a Linux system and networking tools. As many do not have native Linux machine available, you can use virtual machine for running Linux for the exercises. We also use Mininet to emulate different network behaviors in our assignments. Therefore you should start from the setup instructions, that walk through how to install the software needed on this course in a virtual (or native) Linux environment.

We roughly move from the top of the networking stack towards the bottom, with a few sidesteps. For an easy start, we get familiar with basic network programming concepts, first with client socket programming, and then moving to server programming, with techniques to handle concurrency efficiently at the server. We also briefly cover datagram sockets and UDP. Most examples use the Rust programming language, as it is a modern, memory-safe language capable of low-level programming that is gaining popularity in the networking community.

After some programming, we discuss congestion control in the Internet. Even though it is a traditional topic already covered in typical elementary networking course, there are some recent developments and new algorithms worth discussing: for example, these days nearly all systems use the CUBIC congestion control algorithm by default, which is rarely discussed in basic courses. Also the network applies active queue management and explicit congestion notifications, and we discuss also some other new congestion control algorithms that can make use of these techniques.

We will then take a look at the recent developments in web protocols, in particular HTTP. Web techniques on top of the HTTP protocol is not in the scope of this course, but the protocol itself has gone through some significant changes. The new version, HTTP/3 is transmitted over the UDP (instead of TCP), or rather a new transport protocol on top of UDP called QUIC. Among many new features, it separates the congestion control and connection management from managing the multiple streams that compose the web content, and reduces the latency of the communication in different ways. Datagram-based HTTP/3 enables new application possibilities on top of the web protocol stack from Websockets to different forms of proxying and tunneling, challenging the way we have traditionally thought about traditional protocol layering.

We will discuss what happens inside the Linux kernel, and how the Linux processes the network traffic: important data structures, buffering, and how network drivers and offloading is used to optimize the communication efficiency as the performance requirements are constantly increasing. We also discuss the Netfilter framework that can be used for various kinds of modular traffic management and how network namespaces can be used to build virtual networks, for example for the purposes of containerization.

As networks have become more complex to manage with advanced requirements on quality of service and traffic filtering, and cloud datacenters have become common, making the network environment more dynamic, software defined networking has been proposed to manage the complexity. In the SDN model, instead of integrated vendor-specific manually configured hardware, there is a uniform centralized control plane that operates the software-based switches in its domain based on higher-level instructions from network administrators. We will discuss the architectures for software defined networking, along with couple of specific technologies: the P4 language used in SDN switches and network devices and Extended BPF (eBPF) , a way to introduce programmability and user-level interfaces to the network devices and different hook points inside the operating system kernel.

Continuing from SDN, we will discuss datacenter networks that often leverage SDN techniques, and the topologies and architectures typically applied in them. Datacenter networks differ in many ways from the traditional Internet: the network is typically under single administration, but has tight requirements on the data transfer speeds, response times, and reliability. Because of the specific features in datacenters, there are special ways to arrange the network topologies, and the related routing and congestion control solutions are also different, as we will discuss.

Finally, we discuss a couple of very different environments: delay and disruption-tolerant networks, and Internet of Things. Delay-tolerant networks are used in environments where the normal network infrastructure and normal protocol solutions are not applicable, for example because the delays are very long and network connectivity is failing frequently, such as in space. On the other hand, some network environments have hard limitations on available power and computational capacity, and those, too, need specific communication solutions.

Assignments

The assignment descriptions and other possible files needed for assignments are under the assignments folder in this git repository. Some assignments also contain code templates implemented in Rust that can be used to help you to get started with the assignment. You may use the templates or implement your own solution from scratch.

One option is to clone or fork this repository to your local system, after which you can start modifying the provided assignment templates, and maintain your work in a forked personal git repository. This makes it easier to synchronize your modifications between different systems, for example if you want to develop you assignment code in your native system and development tools, but run the code in the virtual Linux guest, that is technically a different machine in your system.

The following assignments are available. Before starting the assignments, you should set up a Mininet-based exercise environment used in the assignments: