This week, the 737 Max was cleared to fly again. According to the Financial Times, the crashes that doomed the 737 Max were caused by software that kept pushing the nose downwards, to the point where pilots were unable to regain control. The original design made decisions based on one sensor alone. The new redesign requires two sensors to agree, and pilot intervention after a defined number of changes. Pilots are now also recommended to do simulator training. The Financial Times claims that simulator training is expensive, so airplane vendors are sometimes hesitant to recommend it.
Image source: Wall Street Journal
Networking is nowhere near this level of autonomy and decision making, well except for routing protocol-based path selection of course. However, as networking inevitably moves in this direction, for good reasons, experiences from other industries are important learning opportunities, whether it be aviation, automobile, or other. From aviation we might consider the problems of making decisions based on one sensor. From automobile we might consider the application of simulation and reinforcement to accelerate learning.
Centralized autonomy has a particular challenge within the Internet culture, as distributed autonomy is central to the ethos of a survivable network.
Autonomy is a good thing. It is a direction networks should go in. As previously stated, the Internet is already autonomous in important ways. The industry should keep pushing forward in this direction whether it is autonomous fault diagnosis, preemptive route changes, or more informed routing (CPU utilization, memory utilization, queue depth / packet drops, etc.).
I suspect that as the industry moves in this direction, the most successful companies will be open and transparent about both the opportunity, the gotchas, and how to design for success. Cultures that bury the truth, realities, and important considerations, eventually crash. Reality eventually catches up. You can live in a social media bubble for a long time. You cannot live in an engineering bubble for as long.
There are no specific companies I have concern about at this time, but human hubris, be it mine, or others, have ample examples through history, whether it is flying too close to the sun, or not keeping the nose up on an airplane. What we do not know, sometimes is more than what we do know.
The future of autonomous networking is bright, though the industry will move through an interim period first, which I will write about next week. This article is not a call to back away from progress, just a reminder to inspire talented engineers to take into consideration important factors, so the number of 737 Max like stories are kept to a minimum.