My first experience with Lemmy was thinking that the UI was beautiful, and lemmy.ml (the first instance I looked at) was asking people not to join because they already had 1500 users and were struggling to scale.
1500 users just doesn’t seem like much, it seems like the type of load you could handle with a Raspberry Pi in a dusty corner.
Are the Lemmy servers struggling to scale because of the federation process / protocols?
Maybe I underestimate how much compute goes into hosting user generated content? Users generate very little text, but uploading pictures takes more space. Users are generating millions of bytes of content and it’s overloading computers that can handle billions of bytes with ease, what happened? Am I missing something here?
Or maybe the code is just inefficient?
Which brings me to the title’s question: Does Lemmy benefit from using Rust? None of the problems I can imagine are related to code execution speed.
If the federation process and protocols are inefficient, then everything is being built on sand. Popular protocols are hard to change. How often does the HTTP protocol change? Never. The language used for the code doesn’t matter in this case.
If the code is just inefficient, well, inefficient Rust is probably slower than efficient Python or JavaScript. Could the complexity of Rust have pushed the devs towards a simpler but less efficient solution that ends up being slower than garbage collected languages? I’m sure this has happened before, but I don’t know anything about the Lemmy code.
Or, again, maybe I’m just underestimating the amount of compute required to support 1500 users sharing a little bit of text and a few images?
“GC overhead” only matter for extreme realtime applications, like emulators, games, drivers, simulators, etc. a 10msec (or even a 100msec) pause in a request processing isn’t gonna even be noticed when your network, database and disk IO are literally orders of magnitude higher. Use Rust for web services if you like the language, comfortable with it, etc. Don’t use it because you think it’ll give you “more performance” or “reduce GC overhead”.
Java, C#, Python, Node, or even PHP as languages will never be your web backend bottleneck. Large scale web services performance tuning is entirely architectural. What caches you keep, how you organize your data, how many network operation does 1 user interaction translate to, stateful vs stateless components etc.
+1, exactly this.
As an aside, “stop the world” GC pauses can affect web server performance in interesting ways. Some web application servers have a perf profile where throughput drops off a cliff as the server approaches max memory load. This is fine, so long as you know what’s happening, and can tune your auto scaling to spin up new servers before you start to hit that threshold. This likely wouldn’t be a reason to not use a particular lang / server, except at the most massive scales.
Meta: Hmmm… replying to kbin.social users appears to be bugged from my instance (lemmy.world).
I’m replying to you instead. It doesn’t change the meaning of my post at least, but we’re definitely experiencing some bugs / growing pains with regards to Lemmy (and particularly lemmy.world).
GC overhead is mostly memory-based too, not CPU-based.
Because modern C++ (and Rust) is almost entirely based around refcount++ and refcount-- (and if refcount==0 then call destructor), the CPU-usage of such calls is surprisingly high in a multithreaded environment. That refcount++ and refcount-- needs to be synchronized between threads (atomics + memory barriers, or lock/unlock), which is slower than people expect.
Even then, C malloc/free isn’t really cheap either. Its just that in C we can do tricks like struct Foo{ char endOfStructTrick[0]; } and store malloc((sizeof(struct Foo)) + 255); or whatever the size of the end-of-struct string is, to collate malloc / frees together and otherwise abuse memory-layouts for faster code.
If you don’t use such tricks, I don’t think that C’s malloc/free is much faster than GC.
Furthermore, Fragmentation is worse in C’s malloc/free land (many GCs can compact and fix fragmentation issues). Once we take into account fragmentation issues, the memory advantage diminishes.
Still, C and C++ almost always seems to use less memory than Java and other GC languages. So the memory-savings are substantial. But CPU-power savings? I don’t think that’s a major concern. Maybe its just CPUs are so much faster today than before that its memory that we practically care about.
Only for things that you specifically want shared between threads – namely this (synchronized refcount) is an
std::sync::Arc. What you want to share really depends on the app; in database-backed web services it’s quite common to have pretty much zero state shared across threads. Multithreaded environment doesn’t imply sharing!The refcount absolutely is shared state across threads.
If Thread#1 thinks the refcount is 5, but Thread#2 thinks the refcount is 0, you’ve got problems.