4 ms·
> I think there are a couple of downsides It's a trade-off. There is no free lunch. Erlang besides high fault tolerance also has soft real-time capabilities a
by rdtsc 14y ago
> I think there are a couple of downsides
It's a trade-off. There is no free lunch. Erlang besides high fault tolerance also has soft real-time capabilities as a goal. Some people don't like that term, but it basically means trying to minimize the latency of the system (even under load).
It is one thing to analyze the system in a sequential manner, compile, optimize the hell out of it, down to assembler, bring up a client, measure and everything looks great. Now it comes time to deploy the system in the field. In the field hundreds of thousands of clients are supposed to connecting and we find that there are so many are dropping because of timeouts and the system is spewing socket errors.
So what happens. The system that was optimized for sequential throughput will not necessarily perform well under concurrent load. The two are often at odds with each other.
So Erlang authors who slowed down the sequential code by inserting those pesky interruptions in the middle of what could be tight C loops, just optimized for a different use case.