WanMil wrote
The profiling data shows that the CPU is most of the time busy with the pbf read and write routines, so I doubt that your disk is the bottleneck.
From my experience profiling is a very time consuming (=> CPU consuming) task. So could it also be that the disk is not the bottleneck while you profile the application but it is the bottleneck while you don't profile it? (It's just a guess...)
This depends on the profiling tool. The overhead for -agentlib:hprof=cpu=samples,depth=20 is very small (a few percent) I think that I can say "IO is no bottleneck" because the first real write function appears in position 56 with just 0.32% : 56 0.32% 74.98% 1154 301828 java.io.FileOutputStream.write The read routine appear in position 3: 3 5.21% 18.39% 18991 300527 java.io.FileInputStream.readBytes but the pbf routines build most of the top time consumers: * 1 7.71% 7.71% 28080 300532 java.util.zip.Inflater.inflateBytes 2 5.47% 13.18% 19936 301089 java.util.zip.Deflater.deflateBytes * 3 5.21% 18.39% 18991 300527 java.io.FileInputStream.readBytes * 4 5.09% 23.48% 18563 301652 java.util.zip.Deflater.deflateBytes* 5 3.42% 26.90% 12471 300813 uk.me.parabola.splitter.BinaryMapParser.parseWays * 6 2.96% 29.86% 10776 301058 java.util.zip.Inflater.inflateBytes* 7 2.75% 32.61% 10025 301108 uk.me.parabola.splitter.WriterGrid.get 8 2.20% 34.82% 8032 300643 crosby.binary.Osmformat$Way$Builder.addRefs 9 2.10% 36.92% 7646 301056 java.io.FileInputStream.readBytes * 10 1.99% 38.91% 7266 300573 crosby.binary.Osmformat$DenseInfo$Builder.mergeFrom 11 1.97% 40.88% 7169 300572 crosby.binary.Osmformat$DenseInfo$Builder.mergeFrom 12 1.93% 42.81% 7030 300577 crosby.binary.Osmformat$DenseNodes$Builder.mergeFrom 13 1.80% 44.61% 6577 300576 crosby.binary.Osmformat$DenseNodes$Builder.mergeFrom 14 1.73% 46.34% 6298 300556 crosby.binary.Osmformat$DenseInfo$Builder.mergeFrom 15 1.65% 47.99% 5998 300553 crosby.binary.Osmformat$DenseInfo$Builder.mergeFrom 16 1.45% 49.43% 5271 300540 crosby.binary.Osmformat$DenseInfo$Builder.mergeFrom * So, I'd say: With pbf format, you don't see much improvement with a faster disk, but with a faster CPU. Ciao, Gerd -- View this message in context: http://gis.19327.n5.nabble.com/splitter-that-generates-problem-list-tp573401... Sent from the Mkgmap Development mailing list archive at Nabble.com.