edge computing is wild to me. why would you create so many new problems for yourself just to get a slightly faster latency? that’s asking for trouble IMO

@kasdeya A lot of it comes down to the fact centralized systems don’t really scale. When your system is really, really large, the complexity imposed by edge compute isn’t actually terribly much.

I think people massively underestimate how important it is to reduce conceptual complexity in any kind of system (software or otherwise)

and by “conceptual complexity” I mean “how many concepts someone needs to keep in mind in order to {work on or troubleshoot} the system”. so that means that (relatively) non-leaky abstractions are okay but leaky abstractions aren’t

@kaylie huh - that really sucks. what I want to say is “they shouldn’t have made such a complex system in the first place - they should have simplified it” but that might be naïve. like maybe some systems are just inherently that complex and there’s nothing to be done about it

@kasdeya yeah like there is a scale of operations at which it can be a useful tool in the belt, but like for even the largest companies we've been at it was pretty overkill

like just set up some regional caches if you need them and call it a day, the percentage of cases where edge architectures make sense are a rounding error

@kasdeya Yesterday’s AWS outage demoed another part of it: DCs like to go down occasionally. Spreading your system across multiple datacenters, while more complex, makes it robust against large and small scale outages.

And when your system is already distributed like that, and at large scale, having those edge nodes starts to pay out as they can simply silently switch what DC they’re chatting with to provide service if something happens, making it less visible to the user that a partial outage began.

@kasdeya Fedi (this platform) is effectively all edge nodes by design and it’s less affected by such physical ongoings as a result.