Open Telecom Platform

I really like the Beam.

I really like OTP, Open Telecom Platform or whatever.

Anyway, the airline stuff for building distributed fault-tolerant systems is really, really good, actually.

So forgetting the, uh, so we probably should start thinking about wrapping up apart from anything else.

Been many bugs and many production fires where someone else's code is your own code, usually.

I think one that I remember from previous company was the usage of stack trace deep into the code and there was like one Erlang OTP release where I think column numbers were added for the first time.

And that had a massive effect, a negative effect on the performance of function.

And this caused when a crash was happening, like cascading effect that was super tricky to figure out.

I think anybody who's done kind of major optimization work will probably tell you the same, that probably like 90, 95% of the ideas that you have for making things faster actually make things worse and make them slower.

But I think one, one thing that, that stands out to me kind of thinking a bit more about that and the kind of the, the story around Jason was handling of, uh, for example, handling of, of Unicode where I thought that, you know, transitioning to using binary pattern matching and like the native UTF eight car matching would, would make things faster, but in, in fact it made them, them slower.

So, so in, in Jason and then now in, in the native OTP Jason library, I re-implementing a custom state machine for parsing Unicode that ended up being faster.

I spent a lot of time investigating why is the native stuff not fast enough? There's certain trade-offs that make it faster in certain situations, but definitely not for this kind of batch validation that I needed for JSON.

brutal.

But the architecture we're looking at offers a completely different path using OTP, the open telecom platform, and a specific abstraction called a gen server, which stands for generic server.

Right.

A gen server gives us the exact encapsulation and separation of concerns of a microservice, but it runs entirely inside our Elixir application.