Hacker News Reader

Comment by AndyKelley | Hacker News Reader

It got the wrong language for my example, a digital audio workstation with peer to peer collaboration.

It said use Go, but Go is garbage collected and that is unacceptable for an app that needs to process real time audio with low latency.

Same thing applies to any networked video game.

View on HN

This is a hard real-time problem. There's a well-known general solution for hard real-time problems: split the problem into a data plane and a control plane, and then program those separately and network the components together (e.g. as two processes communicating over a unix domain socket.)

You can write your GUI in Visual Basic or Node or whatever-the-hell, as long as you've got your audio engine written in something with deterministic latency. If you specify a protocol ahead of time, those can even be separate teams with separate competencies!

Sadly, none of the real-time audio/video/image processing software companies seem to have caught onto this. I guess it's only really "well-known" in the network programming world. (A few game companies seem to have figured it out as well, but not nearly the majority.)

It kind of depends, Go's GC cycle in recent releases is minimized so that no GC cycle will last more than N milliseconds (I forget the specific amount), depending on your specific timing needs, it may be acceptable... You can also write your program so that objects/references are held and GC doesn't happen, depending on your memory constraints and needs (ram is cheap, and systems with 64-256GB are possible relatively inexpensive, compared to developer pay for a month)

Though Rust is probably more appropriate for your needs, in terms of more modern languages.

It's like 30ms which is completely unacceptable.

As for your workarounds, for real time software, these workarounds are too nondeterministic. Real time is one of the primary features, so you need very predictable performance, and that implies having direct control over things like memory.

Like I said, it really depends on your needs... In a game, even, it would depends... a blink of an eye hickup every few minutes in a game is not a huge thing, depending on the game... many games would never even need a large GC cycle on a modern computer. I mean, look at how many fun games were made in Flash...

Yes, if you're doing really real-time, that's one thing... many things aren't, so it depends... that said, again, if you need really real time, a functional approach and rust may be a better option... or a number of other FP languages, for that matter... or C/C++ if you're a masochist, or really anal with your approach (not really trying to offend C/C++ guys, more pointing out that it's rarely the best pragmatic option)

>It's like 30ms which is completely unacceptable.

Not really.

https://twitter.com/brianhatfield/status/692778741567721473

I've seen 10ms mentioned as worst possible scenario, which means not coding with the GC in mind.

There's only one right answer for such systems, and you know it: C/C++. They should have added "High performance desktop app" as an option.

"high performance" is a bit vague. maybe "particularly latency-sensitive" or "requires quasi-deterministic execution time"

I believe you can turn off the garbage collector in go.

Uh. Citation needed.

You can set the environment variable GOGC=off [1]

There doesn't actually seem to be a way to explicitly free memory though so I doubt this is actually used.

[1]: https://golang.org/pkg/runtime/

It's only intended for debugging purposes, near as I can tell (and I'm not sure why anyone would want to do it in the first place beyond debugging):

https://github.com/golang/go/issues/13761#issuecomment-16772...

I stand corrected, there doesn't seem to be a way to manually allocate and de-allocate memory. Looks like there is a switch to turn off GC, but it's intended for debugging.

You might be right, but the language certainly isn't designed around not having garbage collection.

i'm not particularly experienced with go, but it does work directly with pointers iirc. all it lacks is an explicit memory free function, i think. is there more to it?

Reply on news.ycombinator.com