FreeBSD spends 7% of its boot time running a bubblesort on its SYSINITs

Re:FreeBSD spends 7% of its boot time running a bubblesort o... by klyrs on Friday May 19, 2023 @11:33AM (#36003154)

I was mostly joking, but worst-cases include sorted and reverse-sorted lists. I've seen those happen accidentally, in practice, resulting in accidentally-quadratic runtime. This is why, when I taught that class, I made students compare sorting algorithms on real world data sets.

[ Parent | on HN ]

Re:FreeBSD spends 7% of its boot time running a bubblesort o... by cperciva on Friday May 19, 2023 @11:47AM (#36003305)

Sure. In this case though the list doesn't change from boot to boot, or for that matter from system to system; if we hit a worst-case we would notice before the release shipped.

Also, the worst case would be "systems take 2 ms extra time to boot" so it's not exactly a disaster.

[ Parent | on HN ]

Re:FreeBSD spends 7% of its boot time running a bubblesort o... by whimsicalism on Friday May 19, 2023 @11:53AM (#36003390)

> though the list doesn't change from boot to boot, or for that matter from system to system; if we hit a worst-case we would notice before the release shipped.

Wait what - why don't you just sort it at build time then? I'm probably misunderstanding what you are saying or maybe just a premature optimization.

e: Nevermind, read further in the thread and saw you answered [0]

[0]: https://twitter.com/cperciva/status/1659561673278226432

[ Parent | on HN ]

Re:FreeBSD spends 7% of its boot time running a bubblesort o... by cperciva on Friday May 19, 2023 @11:59AM (#36003481)

We might end up doing that. A quicksort takes it from 2 ms down to 40 us though, and doing the sorting at build time would take a lot of linker magic.

(The SYSINITs are records in separate object files which are marked as going into a particular ELF section, so we don't actually have a complete list of them until we finish linking the kernel.)

[ Parent | on HN ]

Re:FreeBSD spends 7% of its boot time running a bubblesort o... by alberth on Friday May 19, 2023 @08:27PM (#36008279)

Any consideration to use Radix sort.

It’s typically the fastest. It’s also a stable sort algorithm.

https://youtu.be/BeoCbJPuvSE

[ Parent | on HN ]

Re:FreeBSD spends 7% of its boot time running a bubblesort o... by sn9 on Friday May 19, 2023 @03:12PM (#36005736)

Isn't the point of quicksort that the randomization of the pivot selection means that you're extremely unlikely to run into a pathological case for the specific pivot chosen?

And perhaps even more unlikely if you permute the input sequence before sorting it.

I'm basing this off of my understanding of Skiena's discussion of quicksort in The Algorithm Design Manual.

[ Parent | on HN ]