tune parallel experiment load params and add resume guards - #161
Merged
Merged
Conversation
fabracht
force-pushed
the
bench-tuned-scripts
branch
from
September 10, 2026 22:28
5ffc7c4 to
ad3dbfb
Compare
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Commits the load parameters the parallel experiment scripts have actually been run with, and makes an interrupted sweep resumable. These tunings previously existed only as uncommitted working-tree edits on a bench VM.
Load parameters
03_throughput_under_loss_v5.shnow runs 16 publisher connections, 8 subscriber connections and--inflight 64(was 4/4 with no inflight). A single subscriber connection receives all delivered traffic through one read loop and one dispatch worker, so too few subscribers caps the measured delivery rate at the load tool rather than the broker. These are the values validated against the broker on GCP, where the publisher host kept CPU headroom while the broker saturated.04_stream_strategies_v5.shand06_resource_overhead_v5.shgain--inflight 64. Their connection counts are unchanged —04deliberately uses a single publisher connection, since the stream strategies differ in how one connection maps topics to streams.03also gains aLOSSES_OVERRIDEhook so a single loss cell can be re-run without editing the script.Resume guards
run_monitored_splitandrun_monitored_pub_onlyskip a datapoint whose final run JSON already exists. A sweep interrupted partway now restarts where it left off instead of redoing completed cells.Validated with
bash -nandshellcheck -S error(clean).group{1,2,3}.envare untouched and keep their placeholders — they hold live GCP addresses and are filled in per session.