[mkgmap-dev] splitter r254

Thu Dec 13 15:36:18 GMT 2012

Felix Hartmann-2 wrote
> oh I meant for splitting europe. Of course for France It stuck to 1 
> pass...  So the performance difference of 12 vs 19-20 minutes was not 
> based on additional passes..
> On 13.12.2012 15:58, GerdP wrote:

OK. I bought a new computer some days ago, so I have now a 4 core machine
with 8GB
and a 64 bit OS. Today I got a new HDD, so I am now copying ....

As I already wrote before, I can split planet with -Xmx7000m
-XX:+UseParallelGC using
--output=xml --overlap=0 --keep-complete=true --max-areas=2048
--max-nodes=1600000
This creates ~ 1500 tiles.

With --output=pbf  or --output=o5m I have to reduce max-areas to something
below 1500, e.g 1024,
else the final distribution phase gets stucked doing GC. 
I assume splitting europe will be no problem 
--output=o5m --overlap=0 --keep-complete=true --max-areas=2048
--max-nodes=1600000

I'll run some tests with different parms. It seems that the performance of
the final (write) phase
depends heavily on the number of threads, available java heap and disk
buffers and fragmentation 
of free space. It is probably needed to find a good compromise between if
you want best performance.

Gerd

--
View this message in context: http://gis.19327.n5.nabble.com/splitter-r254-tp5739717p5740289.html
Sent from the Mkgmap Development mailing list archive at Nabble.com.