Comment by HendrikHensen
5 days ago
> It performs better than Go
Do you have some sources or experiences to share on this topic? I'm very curious. My experience is the complete opposite.
At my previous job there was a Java web application and running in Kubernetes (1 vCPU and 1Gi of memory) was able to deal with at most 80 requests per second, using up almost the full 1 vCPU and ~600-700MiB of memory. That was a bit disappointing, since we were supposed to support at most ~1000rps on this API, 13+ pods is a lot, and we felt "software could do better".
A reactive guru from another team told us we should use a reactive stack, so he came in and rebuilt the app using the reactive stack. Now it was doing about 120rps on one pod (about the same memory usage), which was a good improvement, but still disappointing.
One guy on the team was motivated to rewrite the API in Go as a proof of concept, and we were blown away. With the same 1 vCPU it was now able to handle 400 requests per second, while using ~100-200MiB of memory, and having approximately 20% better response times.
> builds just as fast (at least when not using some popular build tools)
I find this a little bit of a cop-out, because almost everyone is using the popular build tools. And it's quite a chore to build a full application without those popular build tools. With Go, all batteries for building are included.
From your description it sounds like you're comparing an old Java program running on an old JDK with a new Go program running on a new Go runtime. These days Java has virtual threads (similar to Go's goroutines) and doesn't need a "reactive stack", and its optimising compiler and GCs blow Go out of the water without breaking a sweat.
I mean, what is old? How far along are frameworks these days, and what are companies using?
That team was using Java 17. Java 21 was just released and frameworks had no meaningful support for virtual threads whatsoever.
In my experience, most companies using Java are chronically multiple versions behind (e.g. some of my friends still in the Java world are on 11).
Perhaps you can share some sources which prove that Java's performance blows Go out of the water without breaking a sweat? I have not seen any such articles about that in the past 3 years, besides the cherry-picked cases where the JVM JIT can optimize some small algorithmic part of the program on the fly (which is not relevant in most web applications).
I'm happy to stand corrected.
Edit: I found this one using a quick search myself https://medium.com/@mohnisha/java-vs-golang-performance-test... I'll check it out and see if I can reproduce it myself too. Still too bad about that gargantuan memory usage, but I guess memory in the cloud is cheap.
> Java 21 was just released and frameworks had no meaningful support for virtual threads whatsoever.
Spring was ready on day 1, as virtual threads had been an experimental feature since Java 19. Spring Boot added support within a couple months.
> In my experience, most companies using Java are chronically multiple versions behind (e.g. some of my friends still in the Java world are on 11).
And that's on you.
3 replies →
Well, what frameworks were used? Also what is behind the web requests, what were they doing? Because it's comparing mostly frameworks, not languages.
As for the build tools, unfortunately the scene is not the best on Java - but javac itself is indeed fast.
Yeah javac is pretty great. I was doing Kotlin for some time and I was getting very annoying by how much slower and slower running tests and doing full builds became. I thought it was some issue with the JVM ecosystem. Then I started a project with pure Java and I was blown away by how fast it was. Unit tests in my IDE running as fast as I'm used to in Go. And indeed I was surprised by the speed of the compilation step of the build too. Perhaps Kotlin is doing better now, haven't touched it in a few years, but yeah, Java made some great steps there in the past decades.