[OSM-talk] Platform performance update & pladge

[Andy Robinson]

Chatted with Steve and TomC last night. They are both in Copenhagen right
now.

We chatted about two aspects. Chris Schmidt's pledge and the current issues
with the database server.

Dealing with the latter first.

The database server is for some unknown reason running out of memory (I
think it has 1GB physical). This causes the machine to swap using the hard
drive which of course significantly affects performance.

Steve needs to investigate this and he knows it's urgent; however as
currently he is the only one that can sort it we need to be patient. He has
just moved house, is on the road right now and has limited time for other
reasons. This of course is one of the reasons why the OSMF is so important
to him and the rest of us. The result though is that it is unlikely he can
get to look in detail until next week.

In the meantime, and to help the situation, Ben and Mikel have backed off
their TIGER import scripts from the former 1 second cycle to 3 seconds. This
should hopefully ease things a bit. This was implemented yesterday morning
and the stats confirm that the import dropped by over one third yesterday.

On top of this can we make a plea. OSM right now needs to focus upon putting
data in and making sure that users can put data in easily and without errors
and delays. Steve, Tom and I all felt this was important. Therefore if
anyone is running large calls to the database right now to pull data out for
purposes other than editing the data can they hold back those calls until
these performance issues are resolved. If you are running big calls to the
data, can you let one of us know so that we can get a feel for the desired
demand.

This leads onto the second point, the pledge.

Requests for immediate and regular data dumps of all the data need to go to
the back of the request list right now. OSM still needs to focus on data
propagation rather than data export.

Having said that, none of us want to discourage these new rendering and
other great ideas for displaying the OSM data. I'm as keen as anyone to see
the fruits of my labours and hence my immediate take-up of Etienne's
Osmarender work. Right now though there is neither the time nor platform
resources to be able readily to comply with the request. While the money of
course would be most welcome for the project its not the right time for the
fledgling OSM to be focusing on the output when the needs right now are very
much with the input.

Thus my vote goes on reluctantly turning down the pledge while encouraging
all those interested in creating views for the OSM data to make full use of
the monthly planet dump. The June dump will almost certainly be delayed a
little bit but Steve is hopeful that a sans TIGER version will again be
available by early next week.

Cheers,

Andy

Andy Robinson
Andy_J_Robinson at blueyonder.co.uk