feat(bench): PLAIN_ONLY knob — skip causal cells for concurrency sweeps
The c=64 matrix run (job 7dbc1ede) showed plain-mode parity but a 2.2x ka latency penalty; the decisive test is a CONNS sweep, which only needs the plain cells. PLAIN_ONLY=1 skips the two causal cells; the summary already tolerates their absence.
This commit is contained in:
@@ -144,8 +144,10 @@ run_causal() { # $1=mode $2=port
|
||||
# ---- matrix --------------------------------------------------------------
|
||||
run_plain ka "$BASE_PORT"
|
||||
run_plain close "$((BASE_PORT + 1))"
|
||||
if [ "${PLAIN_ONLY:-0}" != 1 ]; then
|
||||
run_causal ka "$((BASE_PORT + 2))"
|
||||
run_causal close "$((BASE_PORT + 3))"
|
||||
fi
|
||||
|
||||
# ---- summary -------------------------------------------------------------
|
||||
say "=== SUMMARY ==="
|
||||
|
||||
Reference in New Issue
Block a user