OSM's future. [Openstreetmap]

[Lars Aronsson]

SteveC wrote:

> * @ 04/02/06 02:53:27 AM lars at aronsson.se wrote:
> > What kind of information can you get out of the tile Squid?  Does 
> > it report the cache/hit ratio, and the min/avg/max processing time 
> > for the cached and non-cached requests?  My impression is that 
> 
> ll:/var/log/squid# grep TCP_MISS access.log.1 | wc
>   17220  172200 2514777
> ll:/var/log/squid# grep TCP_HIT access.log.1 | wc
>   15518  155180 2060270
> ll:/var/log/squid# grep TCP_MEM_HIT access.log.1 | wc
>    4567   45670  623279

I did a simple "time wget" shell command for a single tile URL, 
which apparently is cached by now, and over 24 samples in the last 
48 hours (one sample every 2 hours) I got response times ranging 
from 0.13 to 0.26 seconds, with an average of 0.14 seconds.  
(Yes, this is the elapsed wallclock time, not my local CPU 
cycles.)  This is excellent, even before you consider that 0.03 
seconds of that is the ping roundtrip for 
Linköping-Stockholm-Amsterdam-London. Our problem is not the cache 
hits, but the misses.

Does Squid record the response times for the misses?  That would 
be a useful tool for optimizing the application.

15518+4567 hits to 17220 misses, or a 53 % hit ratio, isn't very 
great for a cache.  The ratio should increase if tiles didn't 
expire after 48 hours.  Do we have the disk to cache more tiles?

Most tiles of the map do not change every 48 hours, or even every 
48 days.  And when they change, 48 hours is too long anyway.  I 
want to see my new drawn maps immediately, so I tend to 
right-click-view-image and manually reload the individual tiles 
that need to be updated.  Perhaps we should just tell map editors 
to do this, and then set the cache never to expire?


-- 
  Lars Aronsson (lars at aronsson.se)
  Aronsson Datateknik - http://aronsson.se