> web benchmarks

https://www.techempower.com/benchmarks/#hw=ph&test=composite...

The Tech Empower benchmarks do seem to reflect general state of Java Web Framework ecosystem with Vert.x being they hyper fast web framework and Spring being way slower.

If you take the standard template for any of these frameworks (both Java and C# and any other language) and you add authentication etc, the real performance will be 5-10% of the numbers reported in those benchmarks. Have a look through some of the weirdness in the implementations it's wild (and sometimes educational). The .NET team especially has done stuff specifically to get faster on those benchmarks.

Could you give me a pointer or two? I wondered about that myself, especially considering the massive improvement from "old" .NET to the core/kestrel based solutions - but a quick browsing a while ago mostly left me astonished how...well, for lack of a better word, banal most of the code was.

Agreed though, lack of all kinds of layers like auth, orm etc. are sadly a drawback of these kinds of benchmarks, if understandable - it would make comparability even trickier and has the danger of the comparison matrix of systems/frameworks/libraries exploding in size. But yeah, would be nice datapoints to have. :)

The custom BufferWriter stuff is pretty neat though, although also not really something most people will reach for. And there is more, like the caching of StringBuilders etc.

But it also doesn't use the actual HTTP server to build headers, but they just dump a string into the socket [2], feels a bit unrealistic to me. In general the BenchmarkApplication class [3] is full of non-standard stuff that you'd normally let the framework handle.

[1] https://github.com/TechEmpower/FrameworkBenchmarks/blob/mast... [2] https://github.com/TechEmpower/FrameworkBenchmarks/blob/mast... [3] https://github.com/TechEmpower/FrameworkBenchmarks/blob/mast...

Puh, yeah, I see what you mean, much the stuff in [2] and [3] is rather...bespoke, especially compared to the minimal and mvc targets. Not really what I'd consider "realistic" as per the benchmark's definition.

But TBH, I wouldn't consider [1] gross, on the contrary - simple, fast, lightweight Razor templating without other MVC (or other external) dependencies isn't that unusual a use case and something I've often thought ASP.NET Core was missing (even Razor Pages feel like overkill if you just want to quickly generate some dynamic HTML).

.NET is perfectly capable of standing on its own, and if there are specific areas that need improvement - this should serve as a push to further improve DB driver implementations and make ASP.NET Core more robust against various feature configurations. It is already much, much faster than Spring which is a good start, but it could be pushed further.

I'd like to note that neither Go nor Java are viable for high-performance programming in a way that C# is. Neither gives you the required low-level access, performance oriented APIs, ability to use zero-cost abstractions and platform control you get to have in .NET. You can get far with both, but not C++/Rust-far the way you can with C#.

Yeah, except if you are working on Web servers the quality of the framework and its supporting libraries is much more important than what code could theoretically achieve. What is the point of being able to 200 mph when you only ever drive up to 30mph.

> Neither gives you the required low-level access, performance oriented APIs, ability to use zero-cost abstractions.

Java is working on high performance abstractions, see Vector API (Simd) and project Valhalla (custom primitive types).

Sure C# has a theoretical leg up (for which it paid dearly by causing backwards incompatibility with reified generics) but most of the libraries don't use low-level access or SIMD optimizations or what not.

If you meant BenchmarksGame, then it's the other way around - Java is most competitive where it relies heavily on GC[0], and loses in other areas which require capability to write a low-level implementation[1] that C# provides.

The only places where there are C calls are pidigts[2] and regex-redux[3] benchmarks, in both of which the Java submissions have to import pre-generated or pre-made bindings to GMP and PCRE2 respectively. As do all other languages, with varying degrees of "preparation".

[0]: https://benchmarksgame-team.pages.debian.net/benchmarksgame/...

[1]: https://benchmarksgame-team.pages.debian.net/benchmarksgame/...

[2]: https://benchmarksgame-team.pages.debian.net/benchmarksgame/...

[3]: https://benchmarksgame-team.pages.debian.net/benchmarksgame/...

Even if you prohibit PCRE2, the .NET submissions using out-of-box Regex engine end up being about 4 times faster than Java.

Surprisingly, even though .NET's BigInteger is known for its inefficiency, it ends up being more memory efficient and marginally faster at pidigits than a Java submission that does not use GMP. The implementations are not line-by-line equivalent so may not be perfectly representative of performance of each BigInt implementation.

My point being - if you look at the submissions closer, the data gives much clearer picture and only supports the argument that C# is a very usable language for solving the tasks one would usually reach for C, C++ or Rust instead.

Sure looks like it's written in Java!

Look at all the programming language implementations that provide big integers by calling out to GMP. Why would it be "cheating" when available to all and done openly? Libraries matter.

2 of 10 (pidigits and regex-redux) allow use of widely available third party libraries — GMP, PCRE, RE2 — because there were language implementations that simply wrapped those libraries.

https://benchmarksgame-team.pages.debian.net/benchmarksgame/...

It’s not that easy. I assume other programs hide the use in macros and libraries, in ways far beyond my simple understanding.

Where there are few enough programs that readers should check that the programs they compare seem appropriate for their purpose.

Why? Did you mean both use intrinsics or both don't?

> Sometimes, you just want to see

As-it-says, look for more-secs less-gz-source-code -- probably less optimised.

But it’s not that easy.

Filters some but maybe not all and maybe filters some wrongly.

If you don't think it's appropriate to compare the pi-digits and regex-redux programs, simply ignore them and compare the other 8!