Shane Hathaway wrote:
>  It's only a hypothesis, but I think the ZEO client code has some 
>  inherent limitations due to its use of asynchronous communication. 
>  Fixing it would require a lot of source code changes.

That hyphothesis further seems to fit with other observations I have had...

Neil Schemenauer wrote:
 >  Our DB fits into the OS page cache.  I don't think IO is a bottleneck.
 >  OProfile results seem to support that theory.

That correlates here to.  Another test I did was to use two different 
clients with cleared caches accessing the same two pages, then they both 
take about as long as each other to retrieve the page, so it implies the 
problem is getting the data from ZEO into the client cache, and not to 
do with the loading and/or efficiency of ZEO getting the object out of 
its data store.

I was thinking though that perhaps this test is a little artificial with 
it running all on the one server, and that it may behave differently 
(less waiting for each other?) if ZEO was on a separate box over a local 
network.  Something else to test!

Thanks,

JB



_______________________________________________
For more information about ZODB, see the ZODB Wiki:
http://www.zope.org/Wikis/ZODB/

ZODB-Dev mailing list  -  ZODB-Dev@[...].org
http://mail.zope.org/mailman/listinfo/zodb-dev