Tony Metzidis compares two WSL stacks using kernel microbenchmarks and a clean GoReleaser compilation. The useful lesson is not the upgrade headline: improvements to one subsystem can look dramatic while barely moving the workload you actually care about.
His measurements, on one Intel desktop with the guest limited to two vCPUs and 4 GB of RAM:
- Sequential memory-copy throughput rose from 7.84 to 12.64 GB/s.
- Process ping-pong latency fell from 20.64 to 18.32 microseconds.
- GoReleaser compilation fell from 191.27 to 183.80 seconds — roughly a 4% wall-clock improvement.
- Reported kernel CPU time dropped much more sharply than elapsed build time. These are different metrics, not additive parts of a wall-clock budget.
The practical takeaway: pair primitive-level benchmarks with an end-to-end task, document resource limits, and control the compiler and module caches. A faster memory path is not a promise of proportionally faster builds.
Treat the article’s explanations involving page reporting and scheduler changes as hypotheses, not isolated causal findings. It compares whole stacks on one machine and does not report repeated-run variability or controlled tests of individual changes. The workload-level result is more useful than the generalized upgrade recommendation.