Skip to content

Add the November benchmark rerun and the CPU-bound comparison with Convex - #64

Merged
abdullah-bl merged 1 commit into
mainfrom
docs/benchmark-rerun
Nov 16, 2026
Merged

abdullah-bl merged 1 commit into
mainfrom
docs/benchmark-rerun

Conversation

@abdullah-bl

Copy link
Copy Markdown
Owner

What changed

Docs only. The Convex comparison was rerun on November 16, 2026, on the same machine and the same Convex image, with onebase built from cc5035c.

  • docs/benchmarks.md: new "Rerun with the cache" section. The five sustained-load tests, one 60-second round each. The October table is unchanged above it.
  • docs/benchmarks.md: new "CPU-bound functions" section. The fixed-work mutation ran on both backends for the first time.
  • Corrected statements. The intro, the dates row, two bullets under "What these tests don't show" and the closing paragraph said the cache hadn't been compared with Convex and that nobody had rerun the CPU test.
  • README.md and benchmarks/convex/README.md: one sentence each that repeated those claims.

What the rerun showed

Test Convex onebase
ping 10,509/s 10,891/s
scan 12,337/s 12,578/s
write 1,085/s 10,540/s
bigdoc 856/s 1,156/s
mixed 2,233/s 10,870/s
  • Scan: the only result that moved. onebase went from 2,467/s to 12,578/s and now matches Convex.
  • CPU-bound mutation: same timings and checksums up to 1 billion iterations. At 4 billion, Convex failed after about 1 s and onebase finished in 3.1 s.
  • Pings during a 1-billion-iteration mutation: Convex's longest wait was 8 ms; on onebase one ping waited 780 ms.

Not included

  • Sustained vector search: it ran after a test that clears the table, so it probably searched an empty one.
  • The table in README.md: it still shows the October scan figures; only the bullet below it gives the rerun numbers.

Each result is a single run. Convex's volume wasn't new (it idled at 494 MiB before the first test), which the page says.

🤖 Generated with Claude Code

…nvex

With the cache, onebase now matches Convex on the repeated scan. The
fixed-work CPU test ran on both backends for the first time.

Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
@abdullah-bl
abdullah-bl merged commit da26936 into main Nov 16, 2026
1 check passed
@abdullah-bl
abdullah-bl deleted the docs/benchmark-rerun branch November 16, 2026 03:44
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants