Amazon blog post in which they describe moving off serverless and to a good old monolith with a thought through architecture:
"Moving our service to a monolith reduced our infrastructure cost by over 90%. It also increased our scaling capabilities."
Looks like without the flow of free money, some teams are coming back to their senses.
@fh I don't disagree.
I guess this article hit too close to home. Currently auditing a company that's stuck in micro service hell on an AWS reference architecture. :P
@klimpong Makes sense - my teams run entire AWS services using either architecture, and they both work fine for what they are doing. It is fun to look at the different tradeoffs :)
I really struggle with the blogpost though - if the initial setup they built already failed at 5% of the predicted load, some of the usual design reviews and discussions must have been horribly wrong, and I am not sure what perspective that gives on the rest on the post..
I'm currently fighting microservices in my current employer!
First thing I said was, "Performance is going to SUCK."
And they said, "But, ... MIRCOSERVICES!!!"
(... entirely missing the point, obviously.)
.
So they got it working, ...
Aaaaand, ...
Performance SUCKS.
Who WouldA Thunk?!?!?! 🙄
.
Like, seriously; I'm "a backend guy," and just *watching* their demo makes me cringe. This thing is *Hard To USE*.
Years ago, I got an interview at Google!!!
I flew out. I did the interviews.
For the whiteboard coding interview, the guy kept saying, "You have unlimited memory."
I looked at the requirements and wrote a "traditional" one-pass small constant memory implementation.
He kept reminding me, "You have unlimited memory available!!!"
...
Well, I'm an old guy.
But I *do know* that *the fastest way* to process data coming from rotating media (or internet steams) is sequential one-pass.
Multi-threading / multi-processing *IS NOT FREE*!!!
Virtual Memory *IS NOT FREE*!!!
Network communication *IS NOT FREE*!!!
So, regardless of whatever "super-impressive ultra-modern technologies" you may have, one-pass with limited working storage is *Freek'n FAST* with all existing hardware.
...
They did not hire me.
.
(But I know what kinds of CPUs they buy. And that tells me that they are actually constrained by processing speed, *regardless* of what they may wish to believe, and what they tell others.)
X
@_lennart maybe a hot take but lambda nanoservices (idk name i made up for running containers per function) turning into something that can use a full cpu makes sense?
idk its one part of the infrastructure so maybe monolith feels like clickbait
@_lennart oh wow. I can imagine the nightmare architecture where the microservice fanboys had been leaving their mark.
To address this, we moved all components into a single process to keep the data transfer within the process memory, which also simplified the orchestration logic.
Yeah, I can imagine.
Lennart Koopmann
Whitney Champion 🍪
Till
Florian Holzhauer
Jeff Grigg
Pumpkin Patch Cable
Elias Mårtenson
kyriii
Aranjedeath
David