Hi Gerd, now we have the problem of how to measure the runtime performance. I have compared unpatched and patched mkgmap. Both versions contain the time output of the patched version. I have used one thread only with 15 european tiles (widely spread over europe) and compare the summarized runtime of the LocationHook only because that's what should be improved by the patch. I have done 4 runs of each version. The mean values are: r2154: 65280 ms for LocationHook, 335325 ms overall runtime patch: 55173 ms for LocationHook, 313082 ms overall runtime diff: 10107 ms improvement Hook, 22243 ms improvement overall The overall improvement is a bit problematic because I would expect that it is close to the LocationHook improvement but its twice. The patch uses less memory and therefore the GC (which probably runs outside the Hook) must do less work. But I am unsure if that's the reason for the good overall improvement. WanMil
Hello WanMil,
I tried it. With small input files I see no change. With larger tiles, it seems to be a bit faster, e.g. runtime decreased 270 to 265 secs.
Gerd
Date: Sat, 31 Dec 2011 01:18:18 +0100 From: wmgcnfg@web.de To: mkgmap-dev@lists.mkgmap.org.uk Subject: [mkgmap-dev] [PATCH v2] LocationHook speedup
I tried to improve the first patch by removing anything not required in the Quadtree and by using a different internal data structure.
I've seen performance improvements but please try and test yourself :-)
The most time is now spend in the creation of the Quadtree. So if you want to search for more performance just start there.
WanMil
Gerds patches inspired me to look for more things that could be improved.
I found that the Quadtree used in the LocationHook is not very optimal. The patch is a first try to increase the performance. The time required for the LocationHook is reduced by 10-50% which is great.
Warning: I haven't checked so far if the results are equal. So maybe there are big bugs in the patch... (and the speedup comes from the poor implementation)
I will do some more tests and optimizations but maybe some of you can have a look on it, test it and comment it.
Have fun! WanMil
_______________________________________________ mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev
_______________________________________________ mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev
_______________________________________________ mkgmap-dev mailing list mkgmap-dev@lists.mkgmap.org.uk http://www.mkgmap.org.uk/mailman/listinfo/mkgmap-dev