Count instructions where timing cannot tell 3% apart (110)

016's second stage passed its criterion: six interleaved pairs against
ee21aec gave TPC-C (RAM, W=1) -0.6% at the median and the bank's
queries 0.0% to +2.2%, no run interfered. The ranks query was slower
all six times, and chasing it showed that on this Pi timing cannot
separate 1-3%: two servers asked in turns disagree with themselves
(ee21aec to e02bfc9 +1.6%, e02bfc9 to 481642f -1.2%, ee21aec to 481642f
+2.8%), and a query whose instructions grew 0.6% ran 3.5% faster.

Instructions do not move so: pcount.c counts the server's user-mode
instructions with perf_event_open (no perf here), and psample.c samples
where they go. ranks had grown 0.4% in instructions (the rest was where
code and data landed), per_branch 3.6% and tables 20 instructions a row,
which 07047c3 halved. The six pairs of 07047c3 were interfered (another
session's build, load 4-12) and are kept apart, unused.

1 files changed, 29 insertions, 0 deletions

diff --git a/research/experiments/025-i64-cost/runs.sh b/research/experiments/025-i64-cost/runs.sh
new file mode 100755
index 0000000..328e11f
--- /dev/null
+++ b/research/experiments/025-i64-cost/runs.sh
@@ -0,0 +1,29 @@
+#!/bin/sh
+# The runs of docs/design/016's second stage, from the repository's root:
+# pure-serve before i64 (9a03e72, built in a scratch worktree) and after (the
+# tree as it is), in turns: TPC-C in RAM (W=1, 64 terminals), then the bank's
+# queries, six rounds, the order flipped every other round. The modules are
+# the same bytes on both sides (aup before and after i64 builds tpcc.aup,
+# bank.aup and the four queries alike), so only the runtime differs. The
+# results go to a directory named for the after commit: runs.sh fc65d66.
+set -e
+out=research/experiments/025-i64-cost/$1
+mkdir -p $out
+work=$(mktemp -d)
+git worktree add -q --detach $work/tree 9a03e72
+make -C $work/tree -s build/pure-serve
+cp $work/tree/build/pure-serve $work/pure-serve-before
+git worktree remove --force $work/tree
+make -s build/pure-serve
+cp build/pure-serve $work/pure-serve-after
+for round in 1 2 3 4 5 6; do
+    first=before second=after
+    if [ $((round % 2)) = 0 ]; then first=after second=before; fi
+    for version in $first $second; do
+        echo "$round $version $(uptime)" >> $out/2026-09-30-load.txt
+        python3 benchmarks/tpcc/run.py --serve $work/pure-serve-$version --warehouses 1 --check 0 --terminals 64 \
+            --sqlite-threads '' --sqlite-group '' --seconds 20 > $out/2026-09-30-tpcc-$version-$round-linux-aarch64.jsonl
+        python3 benchmarks/bank/queries.py --serve $work/pure-serve-$version \
+            > $out/2026-09-30-queries-$version-$round-linux-aarch64.jsonl
+    done
+done