News for hackers. Stuff that matters.
Re:How do you reason about a probabilistic distributed system? by mjb on Friday September 11, 2020 @03:12PM (#24446285)
Very cool post. I like this point particularly:

> In this case we can’t assume, in general, that the thread scheduler will assign each thread processor time with uniform probability. That assumption makes even less sense in a fully distributed system with processes running on separate computers connected by a network.

I would love to see tools in this space mature some more. TLA+ has had a huge influence on how I approach distributed and concurrency problems, and a large influence on the industry in general. Modelling non-determinism (beyond the 'try all paths' approach), and assigning probabilities to outcomes, seems like the large missing piece needed to model real-world systems and protocols.

In systems with limited redundancy, for example, there's always a "transition to a failure state" path which tends to get ignored in modelling with tools like TLA+. That's because these tools can tell you that the whole thing might fail (which is useful), but not the probability that the whole thing would fail (which is significantly more useful).

There are no comments on this comment yet.