Methodology

1. Open Opera mini
2. Disable image loading, select socket connection, save
3. Open this overview page.
4. Save the overview as an offline bookmark.
5. Settings - data counter - clear
6. Return to saved overview
7. Open first test page
8. Write down: Settings - data counter
9. If the test contains a form, fill the input and submit it. Otherwise click on the board. Wait until the result loads.
10. Close test page
11. Subtract from the previous value: Settings - data counter
12. Settings - data counter - clear
13. If you were served an advertisement or if it took more than 2s to load (implying internal retries), redo the test
14. Repeat for each test page
15. Archive rendered obml from app storage

Test results

Bytes in uncompressed obml file, network kilobytes sent and received on initial loading and then on subsequent clicks are noted respectively. Tests prefixed by a "j" indicate that the interface would not be usable without JavaScript, while other tests should also work with server side rendering.

Variance of about +/-100 bytes is normal. Top level navigation is still more costly than interacting with a single page using JavaScript - the latter sends 1kB and receives about 0.5kB less on each click.

Input on a form

1. jxtv 849 2.0+1.4 1.0+1.1, accommodate variable width textarea, string
2. jdi 906 1.9+1.4 1.0+1.1, variable width div, string
3. jxiv 1027 2.0+1.4 0.9+1.3, accommodate variable width writable textarea, string
4. tid 3590 2.0+2.4 0.9+2.1
5. tic 5240 1.9+2.4 1.0+2.6
6. tib 7678 2.0+3.0 0.9+2.6
7. tia 10485 2.1+3.1 2.2+3.5
8. pi 3191 2.3+2.3 3.9+2.4
9. jpi 3147 2.0+2.2 0.9+2.3, submit sent as images
10. xt 2624 2.0+2.2 1.0+1.9, font not monospace aligned in textarea, string
11. jxt 2625 1.9+2.1 1.1+2.0, font not monospace aligned in textarea, string
12. cr 9380 2.0+3.2 0.8+2.7, radio circle and submit sent down as three images
13. jcr 9328 2.0+3.2 0.9+2.7, radio circle and submit sent down as three images
14. jti 4820 2.0+2.8 1.6+2.5, submit sent as images

Clickable

1. jdaj 3031 2.0+1.7 0.9+1.4, lines in string, control via sequence
2. jdao 3254 1.9+2.1 0.9+1.8, lines in string, control via sequence
3. jdas 3253 1.9+2.1 0.9+1.9, lines in string, control via sequence
4. jcoj 4747 2.0+2.0 0.9+1.6, control via pointer
5. jcaf 7409 2.0+1.9 0.8+1.6, control via pointer
6. jcsf 5366 2.0+2.0 0.9+1.7, control via pointer
7. jcsj 5366 2.0+2.0 0.8+1.7
8. ca 5738 1.9+2.0 0.9+1.8
9. jcas 5840 1.9+2.2 0.9+2.1
10. jcos 5082 2.0+2.3 0.9+2.1, control via sequence
11. csp 5377 2.0+2.0, interaction not tested without backend

Utility

1. fontsize 21462
2. sublate - measure gateway state erasure timeout

Unusable

1. xia 3652 2.0+2.4 0.9+2.1, bad render as a cropped pre
2. xib 3690 2.4+2.4 0.9+2.1, bad render as a cropped pre
3. jxi 3645 2.0+2.4 0.9+2.0, bad render as a cropped pre
4. ibt 1007 2.0+1.4, cursor does not appear, can't click
5. ib
6. ib (low): 2.0+3.3 0.9+2.9, downscaled to fit to screen, cursor missing, only the center coordinates sent when clicked
7. ib (medium): 2.0+3.2 1.0+3.0, downscaled to fit to screen, cursor missing, only the center coordinates sent when clicked
8. ib (high): 1.9+2.4 1.0+2.4, scrollable, full resolution, coded as four image slices of viewport size, cursor appears, only the center coordinates sent when clicked
9. ib109 (high) 1269 2.0+2.0 0.8+1.9, a single image is sent with inferior compression, only the center coordinates sent when clicked
10. if exception, possible due to framing HTTP headers on test server
11. jib 3707 2.1+3.7, cursor does not appear
12. jpo 1075 1.9+1.4, cursor does not appear, but pre stored as string?!
13. jpdo 5368, cursor does not appear
14. jtoj 2967 2.0+1.6, cursor appears, not clickable