> * Will tile download be done with PHP (has this been covered in > the benchmarks - compare executing PHP code to extract data from > a large tile with having Apache load the tile directly), or > will we have an Apache module for that?
What about CGI script? It should be much simpler than apache module and it will have access to sendfile() (zero copy), so it should be almost as fast as loading files directly by Apache. > * What transition strategy would we employ - surely we don't want > to delete the current tiles, so they would either have to be > converted to the new format (how long would that take?), or the > tile download code would have to employ a "look for big tile > first, deliver from old data if nothing found" mechanism. I would go for "look for big tileset first". > * Has the "incomplete upload" question been cleared (client uploads > individual tiles that need to be incorporated into an existing big > tile)? Can this be done without a significant performance penalty, > and if not, how frequent is the situation? Creating one big file is so much faster that extra processing needed for merging tilesets shouldn't be significant. Anyway I didn't write any benchmark to prove it. I've analyzed [EMAIL PROTECTED] server log to find out how often this happend. You can find it in one of my previous email. I've also attached csv file showing complete/incomplete ratio to this email. In short, incomplete downloads were quite rare (about 10%), but in last few day incomple uploads were about 30%. Other stuff: It might be useful to include some metadata in aggregated file. I would go for sure for blank land/sea info, as it doesn't add to file size. Other useful info may be user name and version of the client. Layer name and tileset coordinates won't hurt, but we can stay with relaying on tileset filename. > > Bye > Frederik -- Jiri
"from line","complete uploads","incomplete uploads","ratio" 100000,15628,8952,"1,75" 200000,16344,3736,"4,37" 300000,16725,3181,"5,26" 400000,16566,5491,"3,02" 500000,16855,2792,"6,04" 600000,17198,1619,"10,62" 700000,17462,1683,"10,38" 800000,17275,1672,"10,33" 900000,17180,1628,"10,55" 1000000,17606,1766,"9,97" 1100000,16951,1946,"8,71" 1200000,16558,1601,"10,34" 1300000,15816,1497,"10,57" 1400000,17063,2587,"6,6" 1500000,17822,1732,"10,29" 1600000,17006,1718,"9,9" 1700000,17259,1443,"11,96" 1800000,17463,1951,"8,95" 1900000,17823,1527,"11,67" 2000000,16365,2975,"5,5" 2100000,16602,2247,"7,39" 2200000,17454,1753,"9,96" 2300000,16759,2199,"7,62" 2400000,16221,1449,"11,19" 2500000,16520,2786,"5,93" 2600000,17162,2096,"8,19" 2700000,16582,2386,"6,95" 2800000,15731,1374,"11,45" 2900000,17031,2806,"6,07" 3000000,17219,2543,"6,77" 3100000,18200,1268,"14,35" 3200000,17070,1448,"11,79" 3300000,17754,580,"30,61" 3400000,17312,1593,"10,87" 3500000,18501,1156,16 3600000,16936,2618,"6,47" 3700000,17279,1984,"8,71" 3800000,15418,4062,"3,8" 3900000,15577,3561,"4,37" 4000000,16938,1866,"9,08"
_______________________________________________ Tilesathome mailing list [email protected] http://lists.openstreetmap.org/cgi-bin/mailman/listinfo/tilesathome
