Say that prefetching items did not pay, and why (109)

Fetching a found item's cells, a leaf's keys and each level's keys ahead
made finding and reading items 20% faster on 100,000 items in a micro
benchmark, but TPC-C's clean pairs moved 1% at most and finding alone got
48% slower, so it is not in. The owner's finding samples were the descent,
not the fields: a write right after a read copies the leaf whole anyway,
and what costs is walking structures larger than the caches at random.

1 files changed, 26 insertions, 0 deletions

diff --git a/research/experiments/024-finding-items/micro.py b/research/experiments/024-finding-items/micro.py
new file mode 100644
index 0000000..1e80a9e
--- /dev/null
+++ b/research/experiments/024-finding-items/micro.py
@@ -0,0 +1,26 @@
+#!/usr/bin/env python3
+"""bench-seek before and after prefetch.patch, taking turns five times, lists of 100,000 and 400,000 items:
+the median ns of each way and the spread (slowest over fastest), and the load average when it began and ended.
+
+    micro.py DIRECTORY   (holding bench-before and bench-after)"""
+import os
+import statistics
+import subprocess
+import sys
+from pathlib import Path
+
+here = Path(sys.argv[1])
+ways = ['seek', 'seek+3 reads', 'seek+3 reads+3 changes', 'same, a share each 30']
+print('load average at the start:', os.getloadavg())
+for items in (100000, 400000):
+    runs = {'before': [], 'after': []}
+    for _ in range(5):
+        for version in runs:
+            out = subprocess.run([here / f'bench-{version}', str(items), '1500000'], capture_output=True, text=True,
+                                 check=True).stdout.splitlines()[:4]
+            runs[version].append([float(line.split()[-2]) for line in out])
+    for version, measured in runs.items():
+        medians = [statistics.median(m[i] for m in measured) for i in range(4)]
+        spreads = [max(m[i] for m in measured) / min(m[i] for m in measured) for i in range(4)]
+        print(f'{items:>7} {version:<6} ' + '  '.join(f'{w}: {m:.1f} ns (x{s:.2f})' for w, m, s in zip(ways, medians, spreads)))
+print('load average at the end:', os.getloadavg())