<html><head><style type="text/css"><!-- DIV {margin:0px} --></style></head><body><div style="font-family:times new roman, new york, times, serif;font-size:12pt"><div style="font-family: times new roman,new york,times,serif; font-size: 12pt;"><div>I may be misinterpreting, but I really don't appreciate the tone of your email. There is no place for patronization here.<br><br>And I'm sorry if my previous email was short, I had just sent a log email explaining the work I have done on the tile machine and probably should have waited to replied. <br><br><br>I don't control access to the tile machine, but I would recommend that Steve give you access, <br>so you can run performance tests and check the results, and explore solutions.<br><br>I have gone from thinking to knowing (and to doing), and have run tests on Ruby tile generation. I've mentioned this several times in several emails. It would be helpful if I had published the results of these tests, but I simply don't have the
time or inclination to go back into it. I'm convinced Ruby is not the correct choice for tile generation, and will spend my limited time on a replacement rather than more profiling on this portion of the process.<br><br><br>Checked apache's error logs. mod_ruby was reporting a
Logger::ShiftingError in osm/osmlog.rb. Looks like logging had been
added recently, but something was configured incorrectly. I added a "rescue" to osmlog.rb, the errors stopped, and tiles are now generating more consistently. Checked osmlog.rb in svn, and Steve had made the same change a few days ago but hadn't yet pushed it out. I don't know what the problem is with this logging, but now streets are appearing on new tiles. In 48 hours things should be looking better across the map.<br><br>----- Original Message ----<br>From: Lars Aronsson <lars@aronsson.se><br>To: openstreetmap-dev@vr.ucl.ac.uk<br>Sent: Thursday, February 16, 2006 11:17:26 PM<br>Subject: Re: [Openstreetmap-dev] ticket 134<br><br><div>Mikel Maron wrote:<br><br>> Both requests for streets and landsat timed out. The server is <br>> simply overloaded.<br><br>The only thing that is "simply" in this is that the current <br>behaviour is simply unacceptable. My measurements using wget <br>--no-cache indicate that satellite image tiles with streets drawn
<br>take 3 seconds, but when streets are failing the call returns <br>after 2 seconds. If I get a faulty HTTP response (without <br>streets) 2 seconds after the HTTP request was sent, then what is <br>the timeout configured to be? Sometimes I get responses after 6 <br>or 12 seconds - and right now I think that is fine if the streets <br>are drawn. I don't mind the slowness as much as the failure to <br>deliver. Once we get the tile server to deliver, I'm sure we can <br>work on the performance.<br><br>My wget tests were done in the middle of the night, when nobody <br>uses the server (you can prove me wrong by telling me how many <br>tile requests were served every hour of the last week, because I <br>have no idea), and the tile was not downtown London where streets <br>are everywhere, but Swedish countryside.<br><br>Tile generation did work 14 days ago, before Magick was <br>introduced. And it did work for some hours
this Sunday afternoon <br>February 12th. At that time, streets were shown in every tile <br>where they should, and all yellow markers were visible in the edit <br>applet. This enabled me to draw some 50 km of country road and <br>four small villages. But that's the only contribution I have been <br>able to make in the last two weeks.<br><br>> There's one I forgot!! THIS COULD BE CRUCIAL. Apache is also too <br>> heavy duty for this job on the tile server. We'd get much better <br>> performance replacing it with lighttpd. Anyone have experience <br>> with configuring lighttpd?<br><br>No Mikel, this is not it. Apache is fast enough. Together with <br>the Squid and the Linux kernel, Apache is the least suspected <br>component of this system. But the test is easy to do. Let Apache <br>serve a static file, such as a PNG image, and measure (with wget) <br>how fast or slow that
is.<br><br>> I think mainly Ruby takes too long to do all the calculations <br>> and drawing commands.<br><br>How can we move from thinking to knowing?<br><br>One way is to add a CGI parameter &debug=yes to the URL. In <br>streets.rbx this could be used for generating more debug log <br>output. This would not add load to the ordinary requests, only <br>when I want to do my wget tests. And then I would need a way to <br>read that logfile. For example:<br><br>debug = cgi['debug']<br>start = Time.now()<br><br> # rest of streets.rbx here...<br><br>if (debug == 'yes')<br> elapsed = Time.now() - start<br> @@log = Osmlog.instance<br> @@log.log("streets.rbx: Call took " + elapsed + " seconds for "<br> + "bbox = " + bbox + ", "<br> +
"tile_too_big = " + tile_too_big)<br>end<br><br><br>-- <br> Lars Aronsson (lars@aronsson.se)<br> Aronsson Datateknik - <a target="_blank" href="http://aronsson.se">http://aronsson.se</a><br><br>_______________________________________________<br>Openstreetmap-dev mailing list<br>Openstreetmap-dev@vr.ucl.ac.uk<br><a target="_blank" href="http://bat.vr.ucl.ac.uk/cgi-bin/mailman/listinfo/openstreetmap-dev">http://bat.vr.ucl.ac.uk/cgi-bin/mailman/listinfo/openstreetmap-dev</a><br></div></div></div></div></body></html>