Orbi-Fax: second downstairs trial results

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
2026-11-08 19:22:08 +00:00
co-authored by Claude Opus 5.5
parent a0f35f92c8
commit 2843eae7bf
+17
View File
@@ -107,3 +107,20 @@ Any other number shows a short "no page" message, or "coming later" for 101, 300
- **To follow up:**
- The jank on opening and turning pages needs a closer look. Likely candidates are reading every channel's listings in one go, and rebuilding the whole page when only the clock changes.
- The render average is right at the 2 ms target.
## Second downstairs trial, 8 November 2026 (simpler controls)
- **Install:** 0.110.5-orbi-fax-trial2, updated in place (`8ca4d7773edee20b5aa72a81f363950630eba883b0785410c8dcc34d9ea4aad5`). One MainActivity is running and all 2,370 bookings were kept. Profiles and settings are unchanged. Backups and measurements are in `backups/install-0.110.5-orbi-fax-trial2-20261108-1909`.
- **Paul's verdict on the controls:** "feels good".
- **Measurements:**
| Where | Frames 50th / 90th / 99th percentile | Janky | Memory (PSS) |
|---|---|---|---|
| Guide | 9 / 40 / 200 ms | 14% | 116–138 MB |
| Orbi-Fax, 19:12:09 to 19:13:02 | 18 / 26 / 53 ms | 46% | 144–160 MB (Java heap 15–27 MB) |
| First trial, Orbi-Fax | 20 / 40–81 / 150–200 ms | | |
Pages visited: 100, 520, 510, 501 (nine sub-pages), 600, 500 (six sub-pages).
- **Reading the records:** now and next for 38 channels took 202–218 ms on the background thread. The full listings took 411–416 ms.
- **Render time:** a full page averages 1.8 ms, at most 3.0 ms.
- **Still to do:** Orbi-Fax frames now sit just over 16 ms (the 50th percentile is 18 ms), so many still count as janky. The 90th and 99th percentiles are much better than in the first trial. A next step would be drawing the page with a single image upload instead of `setPixels` on each change, and measuring whether the scaled `drawImage` costs more than it should on this TV.