[Openstreetmap-dev] Performance measurment and CSV

[Immanuel Scholz]

Hi,

while I was bored waiting for the big party to start, I did some
performance measurements of the code we currently use to retrieve map
data.

For this, I wrote a small ruby script to generate a dataset for the test
server.
You will find it in sql/make-random-data.rb

Using this data, I test the speed of the various steps within map.rb. This
turned out (the numbers are in seconds):

dao.getnodes: 0.202771
dao.getlines: 1.514142
ox.to_s_pretty: 17.765789
gc: 0.242098

overall (in map.rb): 22.598376
overall (measured at client including transfer): 23.004521

So this means, that almost the whole time (78%) is spent in converting the
data to xml output. The time to transfer the data is practically
irrelevant (2%). Note, that I even did not use the gzipped feature.

WHAT A SURPRISE! (Really, I didn't forecast this.. :-|  )


So its time to either use another XML library, use another platform (no
ruby :-(  ) or reactivate our discussion about CSV.

For the CSV thing: We already have an object modell that supports a rather
linear expression of the data, so defining an transport mechanism should
not be that hard. And btw: Ruby support CSV in one of its standard
libraries.

Any volunteer for a transport format specification? (I'll do it, if you
don't want ;)
Any other comments?

Ciao, Imi.