Marc BrookerDec 19, 2022

I've only really skimmed the ActivityPub protocol, but from what I understand the work done by the whole system scales with:

O(toots * mean followers / mean followers per server)

and the core scaling assumption is that 'mean followers / mean followers per server' scales sub-linearly with 'toots' to stop the end-to-end work being N^2.

Folks who've looked into it, does that sound right?

Show threadMarc BrookerDec 19, 2022

I could definitely believe that property for networks that naturally have a high clustering coefficient (following the terminology from Watts and @stevenstrogatz). In common-sense terms that means small highly-connected communities.

Will Mastodon's communities look like that at scale? Do Twitter's? Or is there something I'm missing in how this protocol scales?

Show threadMarc BrookerDec 19, 2022
The Watts-Strogatz reference is the classic "Collective Dynamics of Small World Networks" https://static.squarespace.com/static/5436e695e4b07f1e91b30155/t/54452561e4b08d9eb2170909/1413817697054/collective-dynamics-of-small-world-networks.pdf
Show threadJared StewartDec 19, 2022
@marcbrooker As you seem uniquely well-read among engineers, would you mind sharing how much time you typically spend reading papers weekly and also how you identify/decide which papers to read?
Show threadMarc Brooker
@Jaredstewart
I wrote a blog post about this a couple years ago: https://brooker.co.za/blog/2020/05/25/reading.html Lately, as I've been deep on some new problems, I've been reading one or two hours a day. Been super valuable to understand our work in context, and find ways to improve it.
Reading Research: A Guide for Software Engineers - Marc's Blog

Dec 19, 2022 at 5:47pmWeb
Show threadjameisonDec 19, 2022
@marcbrooker @Jaredstewart thanks Marc, I appreciate your take
Show threaddefnullDec 20, 2022
@marcbrooker @Jaredstewart Thanks for sharing Marc, I enjoyed reading this blogpost. One of the things I would really like to learn about is - how do you actually organise and write your notes while reading those papers? For example personally I would commonly struggle to recall how certain algorithm work or memorise important aspects of the complicated system described in the paper just a few weeks/months after reading it. And while writing a summary help to some extent often its also becomes a challenge to balance right level of detail, which still does not guaranty memorising. Trying to apply Zettlekasten/Org-roam approach at the moment but I wonder if you have any hints on that matter.