This page “rank” the web servers, then im curious about…
How nodejs can archive 997,327 and rank to 87 and Phoenix 186,774 and rank to 199
Something really strange…
N0x90
danschultzer
Pow Core Team
NobbZ
And many issues more, there are a lot of threads in this forum that talk about the downsides of the Techempower benchmark or about any other.
As I’m on mobile I’ve a hard time to search and link, but I’m pretty sure, someone else will do soon.
peerreynders
jola
You have to understand what you’re measuring. Here’s a screenshot of the same test, Fortunes, but only with JS and Elixir implementations. You’ll see Phoenix far behind a bunch of JS versions.
Now look to the very right of the picture. This column indicates the type of implementation. You can see the JS ones are “Micro” or “Platform”, while Phoenix is “Full”. You’re comparing a full framework realistic implementation with a bunch of stripped-down optimized versions.
No real optimization work has been done on that Phoenix implementation for years too, so there’s no doubt it could be made faster. Although with any change you’d have to think about whether you’re making the implementation less realistic, and then what’s the point? Do a stripped-down version with only Plug or something instead.
I’m not sure where the idea that releases are faster than running with mix comes from, but AFAIK there’s no additional optimization steps for releases, other than preloading modules (which doesn’t matter here since there’s a warmup). I’d be happy to learn otherwise though!
The documentation for mix releases has a good summary of the benefits of releases
mmmrrr
At least one optimization is described in the linked article, if I’m not mistaken:
Code preloading. […] There’s a downside. When you start a new server in production, it may need to load many other modules, causing the first requests to have an unusual spike in response time. Releases run in embedded mode, which loads all available modules upfront, guaranteeing your system is ready to handle requests after booting.
Also: v8 is highly optimized for execution performance and fast startup times. Actually it’s not surprising that it can outperform the BEAM in certain benchmarks and under specific conditions.
Also OT: @NobbZ, your German autocorrect is leaking 
of the Rechenpower benchmark or about any other.
NobbZ
jola
I did mention pre-loading in my post: “other than preloading modules”, and also why it’s not relevant for the benchmark: “which doesn’t matter here since there’s a warmup”.
Yeah, absolutely, V8 will totally win a silly benchmark like Fortunes because it’s just IO and V8 is heavily optimized for it, while the BEAM focuses on other things (like p99).
wolfiton
This article here should prove a lot more interesting for a benchmark
Also the article is discussed here https://news.ycombinator.com/item?id=19089389 and here Benchmarking Go vs Node vs Elixir | Lobsters
mikl
Synthetic benchmarks like these are almost useless.
Unless you’re just serving up static content (and if you are, why use Elixir?), the raw server speed is not really important, since the determining factor for your app performance (speed and scalability both) will be your custom app code, not the framework itself.