Concurrency is a programming abstraction that conceptually allows multiple independent processes to run at the same time using scheduling. For example, running ten programs at once with only a single CPU. Concurrency is effectively CPU scheduling with maybe some other concepts of communication between processes for coordination.
Parallelism is running processes at the same time.
The driving motivation behind concurrency, as far as I can tell, is that it's a concept that can help reason about complex tasks with the added benefit of being able to take advantage of parallelism.
My problem with Pike's advocacy for concurrency, and Golang in particular, is that it's an abstraction that imposes a world view on what's a good programming paradigm and, more practically, how to take advantage of parallelism, both of which have limited utility compared with considering parallelism directly.
My opinion rather than some abstraction that imposes a world view on how to take advantage of parallelism, it's better to use how people actually take advantage of parallelism as a basis for abstraction.
I think Golang is a fine language but we now have 14+ years of hindsight to see how well it's fared and how much it's influenced the compute landscape. In my opinion, the "concurrency not parallelism" misses the simple issue that we want things to go fast. Which has had more influence, Golang's idea of concurrency or taking advantage of GPU parallelism? I think the answer is clearly taking advantage of GPU parallelism.
It seems like we forgot about lightweight fibers for about 30 years and then collectively rediscovered it in about 2010. sure, Java did green threads, but not like m:n stuff.
I sometimes wonder where we would be now if people would have gone "Wow, mr Reppy! concurrentML is so cool!" in 1993.
instead we got pthreads and collective amnesia and later we got go's girly times and channels which are only half way there.
That's because they are very rarely useful. This was true then and it's still true now. There's just not many workloads where it makes sense to need to rapidly launch a thread that doesn't need to do much of anything but does need to exist for a while before terminating.
What is useful is the state machine aspects of things like coroutines or async/await, but those aren't quite fibers and very much aren't M:N threading. A major use of them is in UI where they have strict thread requirements even.
Exactly. When I started getting into go 15 years ago, and 10 years ago or so into elixir, I loved the easy concurrency and tried to use it to the maximum possible. What I discovered is that they're just aren't that many things that you can do concurrently.
Reading many different files or servicing many different network sockets being the exception, but realistically that isn't done very often relative to single threaded stuff.
State machine/management is really what is most useful
Back then you could spend 6 months making something twice as fast, or wait six months for computers to become twice as fast. People obviously did the latter.
It also became a cycle: People write single threaded code -> CPUs/OSs optimize performance/ergonomics for single threaded programs -> People write single threaded code ->...
It wasn't until scaling slowed down that interest/investment in concurrency/parallelism took off.
Yet the average software engineer hasn’t heard or Carl Hewitt, Joe Armstrong, has never programmed in an Actor Programming model, and dont even know what it is.
yeah, i use that often, eg when a video is slow on a website - at least yt-dlp will shard out multiple piece-downloaders, and I can get the video in a few seconds or minutes, and just watch it. Also great for anonymizing sharing videos - download the actual video, share that. Not an url (:
In much the same vein, I rarely actually watch stuff _on_ netflix, through a browser - I watch sped up, and the quality just degrades. Since I pay for it, I feel nothing for downloading a ripped copy to watch it locally :D
After watching the video I can see how go lang makes it easier to write correct concurrent programs. But with AI writing the code these days, it is just as "easy" to write correct concurrent programs in Java (because AI is doing the work). Java's virtual threads are light weight, just like go's routines. Java's LinkedBlockingQueue offers roughly the same functionality as go's channels. I would like to hear from go experts as to why I am wrong. Does go have any inherent advantage if AI is writing the code?
concurrency _is_ parallelism, but for I/O. People often think of parallelism for the case of making something go faster - eg placing two computations in parallel (the definition posed in the video), OR placing two I/O operations in parallel - so this is the keyboard-vs-mouse in the OS, even when you're on one core only; this is multiple web requests in JavaScript, which does not support multi-threading, but 100% does support concurrency for I/O operations - that... badum-tiss! RUN IN PARALLEL.
I get the point of the talk, and it's well interesting, but I think it depends on how one views things.
Not really. They're just separate but related concepts.
E.g. coroutines are a form of concurrency that doesn't have to involve any sort of I/O, you're just taking two logical processes (e.g generating a sequence and consuming it) and abstracting away how they execute relative to each other.
Describing your tasks using the language of concurrency is a requirement for process-based parallelism (multiple CPUs/cores), but data-level parallelism (SIMD) is a form of parallelism that doesn't involve concurrency either.
No, and that's the point of the article. What you are calling parallel w/r/t IO should be called concurrency (conceptually happening at the same time by virtue of being able to interrupt and resume units of work). The reason IO APIs like you've described is concurrent but not necessqrily parallel is because there is no guarantee in the API that they both happen literally simultaneously; I could build a JS runtime that "works" for all the code written against XMLHTTPRequest (ignoring side-effects) but which under the hood only ever makes one HTTP request at a time. And because I can do that, that means JS code is living in a concurrency-only world, even though as an implementation detail most runtimes support parallel execution of those concurrent operations.
> there is no guarantee in the API that they both happen literally simultaneously
There's no actual guarantee in the API that if you spawn multiple threads and call blocking network I/O that those happen literally simultaneously. Maybe the OS has a big mutex on network I/O to serialise them.
Of course, that's not what happens in practice. But neither is it what happens, in practice, to async network APIs called concurrently in one thread. So I don't think that can be the difference between concurrent and parallel.
Parallelism is running processes at the same time.
The driving motivation behind concurrency, as far as I can tell, is that it's a concept that can help reason about complex tasks with the added benefit of being able to take advantage of parallelism.
My problem with Pike's advocacy for concurrency, and Golang in particular, is that it's an abstraction that imposes a world view on what's a good programming paradigm and, more practically, how to take advantage of parallelism, both of which have limited utility compared with considering parallelism directly.
My opinion rather than some abstraction that imposes a world view on how to take advantage of parallelism, it's better to use how people actually take advantage of parallelism as a basis for abstraction.
I think Golang is a fine language but we now have 14+ years of hindsight to see how well it's fared and how much it's influenced the compute landscape. In my opinion, the "concurrency not parallelism" misses the simple issue that we want things to go fast. Which has had more influence, Golang's idea of concurrency or taking advantage of GPU parallelism? I think the answer is clearly taking advantage of GPU parallelism.
I sometimes wonder where we would be now if people would have gone "Wow, mr Reppy! concurrentML is so cool!" in 1993.
instead we got pthreads and collective amnesia and later we got go's girly times and channels which are only half way there.
What is useful is the state machine aspects of things like coroutines or async/await, but those aren't quite fibers and very much aren't M:N threading. A major use of them is in UI where they have strict thread requirements even.
State machine/management is really what is most useful
It also became a cycle: People write single threaded code -> CPUs/OSs optimize performance/ergonomics for single threaded programs -> People write single threaded code ->...
It wasn't until scaling slowed down that interest/investment in concurrency/parallelism took off.
> This video is not rated
> Join vimeo to watch
> Already have an account? Log in
Funnily enough, yt-dlp has no trouble downloading it.
In much the same vein, I rarely actually watch stuff _on_ netflix, through a browser - I watch sped up, and the quality just degrades. Since I pay for it, I feel nothing for downloading a ripped copy to watch it locally :D
concurrency _is_ parallelism, but for I/O. People often think of parallelism for the case of making something go faster - eg placing two computations in parallel (the definition posed in the video), OR placing two I/O operations in parallel - so this is the keyboard-vs-mouse in the OS, even when you're on one core only; this is multiple web requests in JavaScript, which does not support multi-threading, but 100% does support concurrency for I/O operations - that... badum-tiss! RUN IN PARALLEL.
I get the point of the talk, and it's well interesting, but I think it depends on how one views things.
Not really. They're just separate but related concepts.
E.g. coroutines are a form of concurrency that doesn't have to involve any sort of I/O, you're just taking two logical processes (e.g generating a sequence and consuming it) and abstracting away how they execute relative to each other.
Describing your tasks using the language of concurrency is a requirement for process-based parallelism (multiple CPUs/cores), but data-level parallelism (SIMD) is a form of parallelism that doesn't involve concurrency either.
There's no actual guarantee in the API that if you spawn multiple threads and call blocking network I/O that those happen literally simultaneously. Maybe the OS has a big mutex on network I/O to serialise them.
Of course, that's not what happens in practice. But neither is it what happens, in practice, to async network APIs called concurrently in one thread. So I don't think that can be the difference between concurrent and parallel.