Jorge Caballero, MDTwitter's engineering team is just 3 hamsters in an overcoat at this point
There are roughly twice as many engineers working on Mastodon infrastructure as there are Twitter employees (across all functions)
10K servers— let's assume each server requires 0.25 full-time equivalent to maintain: that's 2,500 engineers.
According to this recent article, Twitter's headcount sits at ~1,300...and fewer than 550 are full-time engineers
Ben Aveling@DataDrivenMD
.25 fte/server feels too high. That’s over a day/week. I’d guess, happy to be corrected by someone with actual experience, that’s it’s only a few hours per week, possibly a lot less.
I’d expect (guessing) that it is managing users that would take the bulk of admins time.
.25 fte/server feels too high. That’s over a day/week. I’d guess, happy to be corrected by someone with actual experience, that’s it’s only a few hours per week, possibly a lot less.
I’d expect (guessing) that it is managing users that would take the bulk of admins time.
@DataDrivenMD your original claim is probably true, probably an underestimate, it’s just that most of those engineers won’t be full time.
A Library Socialist@DataDrivenMD that many servers, there's no reason to assume served count maintenance is linear. You're doing some form of automation, hopefully containerized in 2023.
@matty that's why I discounted the value by 75%.
@DataDrivenMD but that doesn't account for it.
If the first server takes 25 engineer hours, but I can reuse the image in a K8s cluster (for example) and it takes .00001 minutes to deploy each additional one, I can run thousands of servers without issue.
@matty Something tells me you haven't tried to scale a Mastodon instance. Either that or you have the world's best Helm chart and aren't sharing with anyone else...and you figured out how to deconstruct a monolithic Ruby on Rails stack so that it's lean and performant at scale...and told nobody else that's running an instance anywhere in the world about it
@DataDrivenMD believe you're discussing twitter here though, not Mastodon. Very different use case for thousands of instances of a company's proprietary server and orchestration versus a federated model, etc.
The point remains - the first one might take thousands of hours, but the scaling isn't linear. Modern scaling almost never is!
jsj@DataDrivenMD Wild if true, what is this based on?
@joseph just added another post to the thread with sources and my back-of-the-envelope calculations
Blaidd Drwg@DataDrivenMD not even that many hamsters
Ken@DataDrivenMD only 3?? I heard the hamsters had taken total control!
D. G. MarshallTHREE hamsters? In this economy?