Prateek's Musings

The biggest ever season of the sport yet concluded last month at Yas Marina Circuit in Abu Dhabi. Red Bull's Max Verstappen who had already confirmed his second consecutive title in Japan saw the chequered flag first, notching up a record fifteenth win of the season. Ferrari's Charles Leclerc finished second in the race and the championship ahead of Red Bull driver Sergio Perez.

A major technical regulation change promised a lot of action on track, but drivers and teams provided even more action off track. Everything has been discussed extensively in the media, so I will stick to talking about the things that made me take notice.

Champions and Challengers

The earlier stages of the season seemed to indicate that Max Verstappen and Charles Leclerc will have a close fight for the Drivers' Championship, and their respective teams will lock horns in the Constructors' Championship. However, a mix of strategy mistakes, reliability issues and damage control to prevent a budget cap breach led to this title challenge ending a lot sooner than expected.

In his first season for Mercedes, George Russell finished in top five in eighteen races and even secured his maiden win at Brazil; a race Sergio Perez would like to forget. Checo did build up a substantial enough gap, but would have hoped for more success to avoid being treated as a support for Max.

Middling Midfield

Alpine and McLaren engaged in a close midfield fight as expected, but both teams were plagued with different issues and therefore ended up quite far from the top three teams. 42 year old "rookie" Fernando Alonso was at the top of his game in a performant Alpine, but ended up with six DNFs thanks to its reliability issues. McLaren's marketing team did a better job than their engineering team, but Lando Norris outdrove the car while overcoming multiple bouts of illness to get the team's only podium finish.

Esteban Ocon's performances didn't stand out much, but somewhat better luck helped him finish ahead of his teammate. On the other hand, Daniel Ricciardo had his worst ever season since replacing Mark Webber at Red Bull in 2014; culminating in him returning to his original team with a fat payout from McLaren for a premature contract termination.

Surprise Switcheroos

Sebastian Vettel's retirement announcement triggered a series of driver transfers for the next season. Fernando Alonso took up Vettel's seat at Aston Martin, and Alpine immediately announced Oscar Piastri as their new driver. However, Oscar publicly denied this move in one of the most famous tweets associated with F1 and decided to replace fellow Australian Daniel Ricciardo.

I understand that, without my agreement, Alpine F1 have put out a press release late this afternoon that I am driving for them next year. This is wrong and I have not signed a contract with Alpine for 2023. I will not be driving for Alpine next year.

— Oscar Piastri (@OscarPiastri) August 2, 2022

Alpine finally signed French driver Pierre Gasly from AlphaTauri ending a long-term association with the Red Bull Academy, and AlphaTauri signed Mercedes reserve driver Nyck de Vries who scored points in his first F1 race filling in for Red Bull Academy graduate Alexander Albon, who decided to stay with Williams for the next season.

While Nicholas Latifi losing his Williams seat to American driver Logan Sergeant after three seasons came as no surprise, Haas shocked some people with their decision to replace Mick Schumacher with Nico Hulkenberg, whose last full season in Formula 1 was in the Renault car in 2019.

Management Mishaps

F1's sporting regulations were already under the scanner due to the sequence of events that happened towards the end of the 2021 Abu Dhabi Grand Prix. Despite this there were noticeable slip-ups in stewarding across the season. It seemed that track limits were not tracked uniformly across circuits, and stewards also ended up giving judgements on a few racing incidents after the race ended, thus preventing drivers and teams from reacting to these incidents.

However, nothing could beat the deployment of a tractor for recovering Carlos Sainz's crashed Ferrari on track, as Pierre Gasly approached at a fast pace unaware of its presence. The memories of the accident in the 2014 Japanese Grand Prix which ended the life of Jules Bianchi are still fresh, and the sport can do a lot better at enforcing protocols which ensure the safety of both the drivers and the marshals.

Cash Crunch

The budget cap became a point of contention for all teams, and the news of Red Bull and Aston Martin breaching it in different ways led to strong reactions throughout the grid. Ferrari apparently had to turn down the engine in the later stages of the season and compromise performance to preserve engines and prevent a potential breach.

I believe budget caps are also being thought of in a similar fashion as restrictions on number of engines, powertrains and other components; which might lead to more teams allowing small breaches instead of trading off performance. Whether imposing budget caps really brings all the teams on a level-playing field is a tough question to answer, but hopefully the structure of these restrictions will become more robust with passing time.

Looking Ahead

The numbers for 2023 represent a bigger follow-up to an already massive season. Next year, the sport will travel to 23 venues across the world including the debut of Las Vegas as the third Grand Prix in USA. The number of sprint races has been increased from three to six. Six teams will field a new driver alongside an existing driver, including two debutants and one veteran. Four teams will undergo leadership change, and teams will learn from the positives and negatives of 2022 to improve their cars with FIA providing some relief for the porpoising issues. The increasing scope of logistics seems to go against the net zero goals of the sport, at least on the surface; but the growing popularity of the sport especially outside Europe is encouraging for a stronger top line and bottom line.

One of the first big problems that many new Linux users encounter is when they break their system while trying to add or remove applications, when they don't understand the impact on other installed dependencies. A popular example of this is tech YouTuber Linus Sebastian breaking his system while trying to install Steam.

However, with increasing Linux adoption for personal desktops and workstations, there has been an advent of new Linux distributions which isolate the base system from user-installed applications to eliminate breakage and create repeatable (if not reproducible) systems. In this post, I try to understand and explain three popular Linux distributions which champion immutability and reproducibility using two different approaches.

Application Containerisation

Docker and Kubernetes, industry-standard tools for creation and management of containerised applications, have made the concept mainstream in the application development space. Red Hat and SUSE brought these principles to operating systems with the Project Atomic and openSUSE Kubic respectively, which have now been retired and replaced by Fedora CoreOS and openSUSE MicroOS respectively. These systems have also been adapted for desktops and workstations.

Fedora Silverblue

Silverblue was first presented as an official spin of Fedora Workstation in October 2018 with the release of Fedora 29. While both distributions are identical in appearance, the distinction between Silverblue and Workstation can be explained in three parts - the base system, GUI applications and command-line applications.

The base system uses rpm-ostree, a packaging system created by combining libostree with Fedora's package manager DNF. The system update mechanism is atomic, i.e. a new image is generated and applied. This allows users to choose from multiple images to boot into, and also rollback easily to the last working state in case of any failure during updates. The base system is intended to be identical by default in terms of packages present, but Silverblue also provides the flexibility of layering packages over the base system.

For installing GUI applications, Silverblue recommends Flatpak, a universal package manager which creates a sandbox comprising of the installed applications and their dependencies separate from the base system. These applications can be installed from the Fedora Flatpak registry or an abridged version of Flathub, the official advertised app store.

For command-line applications, Silverblue comes pre-installed with toolbox based on Podman, a drop-in replacement for Docker. Users can use OCI images for a minimal base to install command-line applications, and even GUI applications (although Flatpak is recommended). Fedora provides official images, and the community has created unofficial images for other popular distributions like Arch Linux and Ubuntu.

openSUSE MicroOS

The architecture (for lack of a better word) of Silverblue and MicroOS is quite similar, but the implementation is somewhat different in all three parts.

The base system of MicroOS uses their default package manager zypper under the hood but delivers atomic updates and rollback functionality by creating snapshots of the BTRFS subvolumes using snapper. Another difference is the absence of layering, thus making the base images 100% identical across deployments.

Users have complete unfettered access to Flathub for installing Flatpak on MicroOS, unlike Silverblue which provides the Fedora Flatpak registry and a curated version of the Flathub, thus limiting the application availability for the latter.

And finally for command-line containerisation tools, users also have the alternative of using distrobox which works with both Podman and Docker and provides images for more distributions, thereby seems to be less limited than toolbox.

openSUSE MicroOS is still deemed as alpha software, but it seems to edge out Silverblue with the use of standard tools like BTRFS and snapper, unfettered Flathub access and the out-of-box availability of distrobox.

Nix and NixOS

While working on his Ph.D at Utrecht University in the Netherlands (read his thesis), Eelco Dolstra created the Nix language for the cross-platform Nix package manager, which provides a way to create reproducible builds of packages and by extension, as evidenced by NixOS.

NixOS provides the same features as the aforementioned distributions, but manages them in a very different manner. It provides an immutable root file system by replacing all global system directories with the /nix/store directory which contains all the binaries and configurations for various system components in their own directories. This aids features like installing multiple package versions, atomic upgrades and rollbacks and rebasing based on changes made to the configuration.nix file, which can be used for deploying reproducible builds across multiple machines.

Apart from NixOS, the Nix language and package manager has also inspired a number of projects which take tools like Docker and Kubernetes to another level. The package repository nixpkgs contains over 40,000 packages built for Linux and MacOS. However, NixOS is meant for the more advanced users and comes with a steep learning curve involving understanding this domain-specific expression language.

Final Thoughts

Linux and Unix based operating systems based on the immutability principle are making the experience more user-friendly for non-tech inclined users who want a just works system and advanced users who desire granular control over the system without the risk of data and other losses. I personally have been using Fedora Silverblue for the last three months and it has been an Adwaita experience. I am getting to learn more thanks to people like Jorge Castro who writes almost exclusively about these systems on his website.

Thanks to minor hardware upgrades to my old laptop, I have gotten the confidence to venture a little further into the Linux space. I haven't successfully installed Arch Linux in the traditional way yet, nor have I moved to a window manager from a desktop environment. But I believe I can identify things which could go wrong and avoid them.

Before the hardware upgrade, despite using different distributions at different points in time, I mostly used the XFCE desktop sans the "goodies", i.e. the additional plugins and applications packaged with the base system. Using that as the foundation, I decided to conduct a small, non-meaningful experiment to understand how distributions with the same foundation handle package management differently. The problem statement was pretty simple -

Install four Debian-based distributions using the non-GUI installer for the base system and the terminal for a minimal XFCE desktop; and compare the package count and cold boot RAM usage.

Choosing The Distributions

Debian is one of the oldest actively developed Linux distributions, and has been the most popular choice for first-order and second-order derivative distributions which could be used both for servers and desktops.

Ubuntu might probably be the first Linux distribution installed by new Linux users at least in the last decade. While Debian is a community project, Ubuntu is commercial-first and has been partnering with big tech organisations like Microsoft to improve cross-platform compatibility.

Devuan is a Debian derivative which does not use systemd, the most popular init system on Linux; and offers alternatives like sysvinit, openrc and runit for the same.

Trisquel GNU/Linux is an Ubuntu-based libre distribution with completely open source software, i.e. it removes all proprietary blobs of code from the kernel and other packages (if any). As a result, it is one of the few Linux distributions approved by the Free Software Foundation.

Installation Process

There were two parts of the installation process - the base installation and the minimal XFCE installation. For the first part, I decided to use the net install ISO images which come with the TUI menu-based Debian installer. These images were readily available for Debian 11 Bullseye, Devuan 4 Chimaera and Trisquel 10 LTS Nabia; however, Ubuntu had a different process altogether which led to a discovery.

Since April 2020, Ubuntu offer their own server install images for setting up the base installation. Prior to this, they released unofficial net install images with the Debian installer. I managed to locate the image for 18.04 LTS Bionic Beaver, used it for the base installation and then upgraded first to 20.04 LTS Focal Fossa and then to 22.04 LTS Jammy Jellyfish.

For the second part, I discovered a Github repository with an installation script for a minimal XFCE desktop. I installed the packages present in the script along with a few others I use to setup the desktop on all four distros.

Findings

The Debian installation had the lowest package count and second lowest cold boot RAM usage among the four distributions, while Devuan switched places with Debian in both categories. Both installations also took the least time, with the only additional step of selecting the init system in the Devuan installation process.

Ubuntu took the longest time to install because of the two upgrades, but I was able to get a system completely devoid of snap packages. The installation was joint third in terms of package count and third in cold boot RAM usage.

Due to package dependency management differences, my Trisquel installation ended up with GNOME components as well which I had to uninstall. Once those were removed, the resultant system had the same number of packages as Ubuntu and the highest cold boot RAM usage.

Final Thoughts

Package count and RAM usage don't really mean much unless there are severe storage and RAM limitations, but using individual packages rather than metapackages can help preventing the installation of packages and dependencies which don't get in the way of your applications, and reduce the possibilities of borking the system.

Also, libre distributions don't really make much sense at this moment. In fact, most users wanting to install them on hardware from mainstream hardware manufacturers will fail as most firmware tends to be proprietary. However, this speaks more to the manufacturers' apathy for open source rather than the distributions themselves.

During a farewell event organised for me by a team I worked with, one of my team members stayed that they appreciated that I asked a lot of questions. I guess the questions started during my first job, when problem solving became essential for earning bread. At IIMA, questions were a device for driving class discussion in the academic sphere. During placement preparation, we were told about the importance of asking clarifying questions to define the scope of problems.

Every decision we make is an outcome of a problem solving exercise, and each problem is essentially a question. An optimised question could unlock the potential for a much better outcome, therefore I believe it is imperative to work on asking better questions.

Use The Internet

The internet is the biggest source of information and knowledge available to us today. The incentive of ad revenue has driven the creation of websites with relevant (looking?) information in concise form accessible to everyone with a smartphone and working internet connection. Traditional information and knowledge sources have also realised the potential of this medium, so our dictionaries and encyclopedia have also moved to the cloud. An individual cannot compete with the capacity, speed and accuracy of the algorithm of a global search engine. So why not check if the answer is directly available on the internet?

Too Big or Too Small?

If I had to look at the entire lifecycle of a question I asked someone, it would start with me formulating the question in my mind, followed by expressing it in some form, then by the other person receiving the question, processing it, formulating their response and expressing it, and ending with me receiving and processing the response. I might have a lot of time to formulate the question or process the response, but the time available for the intermediate activities is certainly limited. You also have no control over the time that the responder could take! Hence, it is essential that we work on optimising our part of this lifecycle.

Is It Clear?

Communication is both implicit and explicit, and the implicit part can add a lot of processing overload because it is usually open to interpretation. Therefore, unless you are posing a rhetorical question or being sarcastic, an important consideration is making the question as clear as possible. A clear question could lead to a quicker and more valuable response contributing to a more effective decision with a better outcome.

Ask The Right Person

Similarities in experiences reduce as the underlying factors change. E.g. my experience in the classroom during a PGP1 course at IIMA would match a lot closely with someone from my section, a little less with someone from another section, lesser when compared with an immediate senior or junior and a lot less when compared with someone five years senior or junior. Asking someone with an ongoing or recent experience of a setting could lead to getting more valuable information, and thereby ultimately improve the outcomes of your decisions.

Final Thoughts

Most of us remember all our teachers asking if we have doubts, pushing us to ask questions. That's because we are rarely able to acknowledge that we have a lot of unanswered questions in our mind. If we are able to identify the existence of such questions, refine them, pose them to the right people and process the answers in the intended form, our decisions stand to improve vastly. Hope all of us are able to use our questions well!