What 100,000 concurrent sandboxes has taught us so far

▤ Summary

The team is refining its 100,000-sandbox benchmark because early test designs measured the test harness more than the underlying provider. It moved from a single-VM approach to a sharded architecture so the workload better reflects real high-concurrency behavior. It also changed the benchmark to keep sandboxes alive simultaneously, which lets it measure true sustained concurrency instead of short-lived burst throughput. The update adds better logging and a dual-store data pipeline so the team can debug failures and preserve raw results across a very large distributed run. The final benchmark results are now delayed until June 17 while the team continues tuning the methodology.

▦ Linked Companies