I created a small blog post about compilation speeds, binary sizes, and request processing performance of Java Spring, D Vibe.d compiled with debug/release mode and DMD/LDC compilers.

https://www.funnel-labs.io/2022/06/17/language-choice-and-performance-java-and-d/

The interesting thing is that a web service compiled with both DMD and LDC in debug mode has very low throughput, much worse than Java, however, D benefits from faster compilation times.

When compiling in release mode, both DMD and LDC have about 80% to 100% higher performance than Java. The DMD and LDC versions also run in significantly less RAM than Java as well (about a 95% reduction in both resident and virtual memory).

A few other interesting observations are that a web-server compiled with LDC takes up about 10x more virtual RAM compared to the DMD version.

The final fascinating observation is this: despite the longer initial compilation times when compiling in release mode, both DMD and LDC have faster iterative compilations in release mode than even debug mode. My suspicion is that the smaller file sizes that result from compiling in release mode result in overall faster incremental compilation. It makes me wonder if larger companies could benefit strongly from even doing release mode compilation, but keeping compiled artifacts in a shared server, following a model similar to what Bazel can offer.

On Friday, 17 June 2022 at 21:37:52 UTC, Vijay Nayar wrote:
...

Great data Vijay, thanks! Benchmarks are always nice to have.

Certainly could work, but we'd need far better tooling.
Too, most compilations are sure to be in debug mode, and already fast. Release compilations are relatively rare-er and if it takes longer, no big deal? I guess I'm not working on projects big enough that it matters -- our release builds still complete in under 60 sec.