This is great news.  This is one of the reasons why I want to get my lab here set up btw. 
Once I have that up and running I can expose some servers for us to use for testing
these things and keeping the server hardware used constant.


On 3/16/07, Emmanuel Lecharny <> wrote:
Hi gang,

I have done some more preliminary tests, after having introduced some
more optimizations in the 1.5 server :
- serialization of entries is now handled directly instead of relying on
default serialization.

The test is the same : 10K searches on my poor desktop (one 3Ghz CPU,
server launched with 512M, no tuning at all - out of the box server.xml - )
To be very clear : these tests where run in order to see what kind of
performance increase we could obtain from those modifications.

Here are the results :
ADS 1.0.2 SNAPSHOT : a maximum of 250 req/s
ADS 1.5.0 SNAPSHOT : a maximum of 135 req/s
ADS 1.5.0 tuned : a maximum of 435 req/s

This is a 75% improvment above the 1.0.2 version, which is quite
impressive. Note that the injector and the server where running on the
same machine, thus limitating any conclusion we can make regarding the
real performance of the server.

I will conduct some more tests in a few weeks, with more powerfull
servers an with an injector, to get more realistic numbers. Any help
would be welcomed (for instance, ssh access to some computer somwhere...
I don't have a 8 CPU server with 4 injectors at home ;)

Emmanuel Lécharny