[OSM-dev] TIGER data show OSM getting slower??

Tom Hughes tom at compton.nu
Sun Sep 16 23:01:07 BST 2007


In message <200709170716310093.0D0F4D6F at smtp.nsw.exemail.com.au>
          "Brent Easton" <b.easton at exemail.com.au> wrote:

> > So, is TIGER+AND causing too much load to the OSM infrastructure right
> > now?  AND will be done soon, but TIGER certainly won't be.  Plus, we're
> > not even _converging_ on a time when the upload will finish.  Should I
> > slow the upload down a bit?
> 
> I don't know if is due to TIGER/AND or what, but over the last few weeks, I
> have found the Quality of Service provided by the API for uploading mapping
> data to be appalling (Using JOSM). I've been doing some coastal trace
> mapping when I have spare time at different times of the day and as a
> percentage of time I would say:
> 
>  50% - slow, but satisfactory
>  25% - slow, unsatisfactory
>  15% - extremely slow, almost unusable
>  10% - Non existent
> 
> 1 in 4 times I sit down to map, I find the system is pretty much unusable.
> Not good. If you want casual users to have a good experience and spend time
> mapping, the upload API has to be a focus for quality control.

I think you must be very unfortunate in your choice of mapping times
then - until about a week ago performance had been quite reasonable
for several weeks.

The start of last week was quite bad, but once gps_points was sorted
out midweek it improved significantly until today when it did struggle
under the load at times - weekends are unfortunately a busy time in
general.

Uploads are generally fine except when there are too many people trying
to download which can lead to a complete lockup for short periods as
downloads are gradually getting slower and slower as the size of the
database increases and a download that takes several minutes hogs a
process that could otherwise process dozens of uploads.

The solution is quadtiling but we need to arrange such API downtime to
convert the node tables. I was hoping to do that at the same time as the
move to API 0.5 but we might have to do it earlier than that as things
are starting to reach a critical level as today has shown.

Tom

-- 
Tom Hughes (tom at compton.nu)
http://www.compton.nu/




More information about the dev mailing list