On Fri, Dec 05, 2008 at 08:35:23PM -0500, Milenko wrote: > I can confirm this. I went ahead and mounted a ram-drive using tmpfs. Then > created a new tablespace located there. I changed map.fcgi to use > temp_tablespaces instead of default_tablespace and pointed it at the new > tablespace. Pgsql is using the space, but I see no difference in > performance. When monitoring the disk-based tablespace, it doesn't appear > pgsql is even storing files there, so the kernel must be caching them > automagically. > > I guess the only performance option is to throw more discs in the array. I > hoped the 4 10k rpm velociraptors would hold up better than they are. > Iostat shows 250 - 300% i/o wait during heavy load with 7 - 8 instances. > Everything else looks good - the system is just sitting waiting for the > discs. > > Maybe 8 of the new Intel SSDs in RAID0 would speed things up... :)
My guess is that the only solution is ram, ram and more ram. The problem
with the node index is basically that the width is very small and thus
the index is nearly as large as the table itself - IIRC its about
11-12GByte. So when you have less memory than that every access is going
to hit disk very early - namely trying to get nodes in a bbox via the
gist index. My guess would be that you'll see good performance gain up to
10-20GByte memory and then once you have the full database in memory but
thats out of the scope of most of us.
Flo
--
Florian Lohoff [EMAIL PROTECTED] +49-171-2280134
Those who would give up a little freedom to get a little
security shall soon have neither - Benjamin Franklin
signature.asc
Description: Digital signature
_______________________________________________ Tilesathome mailing list [email protected] http://lists.openstreetmap.org/listinfo/tilesathome
