spaetz wrote: > Hi all, thanks for bearing with the recent outages on the [EMAIL PROTECTED] > server, it was overloaded for several reasons (e.g. having OSMXAPI and [EMAIL > PROTECTED] on the same server doesn't help as one is very busy when the other > is). One of the big problems proved that the upload slots were taking up all > apache slots quite quickly and many of the uploads took between 5-8 mintes. I > don't know whether that is because people are running [EMAIL PROTECTED] from > slow home ADSL connections with crappy upload times. Thanks go to Firefishy > (Grant Slater) for helping debug the apache server.
I could imagine that quite a few people upload through ADSL, which would easily take 5 - 10 minutes on the more complex tilesets. I know that I do. I could stop if it is seen as a problem though. Hopefully the move to the colour quantized tiles will have reduced the average tileset size by quite a bit and thus the time it takes to upload. Some of the load on XAPI is probably caused by [EMAIL PROTECTED] clients? Does this fallback make sense at the moment in this case? I feel that nearly always when the main API has no data (i.e. times out), the same happens to the XAPI. So it might only add extra load onto the XAPI without gaining much? I don't have any numbers to back this up, so I don't really know if this would help or harm. > > I have now installed a lighthttp instance with 7 slots and all uploads will > be redirected to using that. So from now on, uploads should not prevent us > from doing tile serving. Let's see if this helps already. Although this is a great idea in general and I think it makes complete sense to split those two as well as possible, since then I can't upload any tiles any more. I suspect it is caused by a stupid firewall that blocks port 81 on our university network. Do more people have this problem? > > Another issue - and that is where you can help - is that currently all tiles > are served via mod_python. While this works fine and reasonably fast, it > consumes quite a bit of memory. I'd appreciate if someone could sit down and > would write a C-based (or whatever high performance, low overhead lingo you > prefer) fastcgi that does tileserving. All the python code is in those 2 > files: I had started writing an apache module in C for the tileset stuff before I heard that Spaetz rewrote the entire server code. If it is helpful, I would be happy to continue on it again and adapt it to the new server layout. It would probably take a couple of days though till I find the time to finish of the testing. Please let me know if other people are working on it or if the memory consumption is no longer an issue to prevent duplicate effort. > > http://svn.openstreetmap.org/sites/other/tilesAtHome_tahngo/tah_intern/serve_tiles.py > http://svn.openstreetmap.org/sites/other/tilesAtHome_tahngo/tah_intern/Tile.py > > And I would be happy to give assistance to those needing it. > > Thanks, > spaetz Kai > > _______________________________________________ > Tilesathome mailing list > [email protected] > http://lists.openstreetmap.org/cgi-bin/mailman/listinfo/tilesathome _______________________________________________ Tilesathome mailing list [email protected] http://lists.openstreetmap.org/cgi-bin/mailman/listinfo/tilesathome
