rerun the benchmark suite and put the cross-language state up front - #619
Merged
Conversation
full rerun on 2026-07-29, after the green default, the blocking-call work, and the ownership sweep. the scoreboard at the top of docs/performance.md now carries the fresh numbers, split into coordination and services-and-compute, and a new section lays out what the july hardening changed as before/after measurements rather than adjectives — the worker-stall fixes, the tls survivability, and the six leak shapes the growth gate now pins. the honest movements both ways: chan_fanout green improved to ~133ms (level with rust, ~1.8x go, still bimodal on placement); the catalog workload runs ~4x faster than go's version and is the strongest row; the os-thread spawn rss collapsed 174mb -> 3.5mb as a side effect of containers releasing on eviction; event_ledger and std_pipeline drifted a few percent slower as hot paths gained the releases they had been skipping; and a minimal go spawn twin shows go clearly ahead on raw spawn/await, which the old table understated. the http server row records what it exists to show — a third of go's throughput at byte-flat rss where go grows. bench/README.md tables updated in place with the rerun dates kept next to the numbers they belong to.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
summary
full benchmark rerun on 2026-07-29 — event ledger, chan_fanout, spawn,
std_pipeline, the grpc unary echo, the catalog workload, and the http server
under wrk — after the green default flip, the blocking-call work, and the
ownership sweep. docs/performance.md opens with a fresh two-table scoreboard
(coordination, services and compute) and a new "july 2026 hardening, in
numbers" section that states each change as a before/after measurement: the
three worker-stall fixes, the tls one-packet-kill fix, and the six leak
shapes the growth gate pins, plus the cost side (the file-i/o handoff and a
few percent on the compute rows).
notable movements, both directions: green chan_fanout ~133ms, level with
rust and ~1.8x go; the catalog workload ~4x faster than go; os-thread spawn
rss collapsed 174mb to 3.5mb as a side effect of eviction releases; a
minimal go spawn twin shows go ~7x ahead on raw spawn/await, which the old
table understated; event_ledger and std_pipeline a few percent slower with
the leak fixes in their hot paths.
what was tested
every number in the diff was measured on this machine during the rerun:
interleaved runs with medians (7 rounds for the fan-out, 5 elsewhere), the
grpc client prebuilt so compile time stays out of the measurement, digests
and checksums verified where the benchmark defines them.
make docsite-checkpasses.