From 68dca986ee7ff7d729a15565e2e07127d5b5d1a4 Mon Sep 17 00:00:00 2001 From: eunhhu Date: Wed, 30 Sep 2026 13:52:30 +0100 Subject: 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. --- .../024-finding-items/2026-09-30-micro-linux-aarch64.txt | 6 ++++++ 1 file changed, 6 insertions(+) create mode 100644 research/experiments/024-finding-items/2026-09-30-micro-linux-aarch64.txt (limited to 'research/experiments/024-finding-items/2026-09-30-micro-linux-aarch64.txt') diff --git a/research/experiments/024-finding-items/2026-09-30-micro-linux-aarch64.txt b/research/experiments/024-finding-items/2026-09-30-micro-linux-aarch64.txt new file mode 100644 index 0000000..df7d999 --- /dev/null +++ b/research/experiments/024-finding-items/2026-09-30-micro-linux-aarch64.txt @@ -0,0 +1,6 @@ +load average at the start: (3.32470703125, 3.7529296875, 3.55712890625) + 100000 before seek: 125.4 ns (x1.69) seek+3 reads: 339.3 ns (x1.36) seek+3 reads+3 changes: 433.7 ns (x1.33) same, a share each 30: 681.5 ns (x1.15) + 100000 after seek: 185.9 ns (x1.10) seek+3 reads: 270.8 ns (x1.08) seek+3 reads+3 changes: 360.1 ns (x1.07) same, a share each 30: 597.2 ns (x1.12) + 400000 before seek: 212.6 ns (x1.10) seek+3 reads: 384.7 ns (x1.02) seek+3 reads+3 changes: 505.4 ns (x1.12) same, a share each 30: 855.1 ns (x1.22) + 400000 after seek: 261.8 ns (x1.55) seek+3 reads: 348.8 ns (x1.63) seek+3 reads+3 changes: 470.6 ns (x1.18) same, a share each 30: 801.6 ns (x1.51) +load average at the end: (3.677734375, 3.7158203125, 3.5546875) -- cgit v1.2.3