performance generally

["Andrew C. Oliver" <ac...@apache.org>]

As is probably obvious I'm working on TBird integration.  In comparing
opends (https://opends.dev.java.net/) with the default configuration and
example data the return results are much faster than ADS 1.0 (as noted
1.5 is not working ATM
https://issues.apache.org/jira/browse/DIRSERVER-927).  Is it just a
configuration, storage thing or index thing or a real differential?

I followed the tutorial (as mentioned previously) then loaded the OpenDS
example data into ApacheDS via the apacheds eclipse plugin thingy ldif
export/import.  then I used a regexp

For whatever reason a few entries didn't get copied to ApacheDS
(probably I missed a limit to 1000 or something).  However that should
have only affected the results in the other direction.

Searches on general things like A or B and specific things like Bonnar
or "Bonnar Billoteau" show the same perceivable performance difference.
  This bears out in both the GUI and Wireshark as to returned result
timings.  ApacheDS handling thing connection better/faster but OpenDS
handling the response in the sub-sub ms rounding error.

Thoughts?

)

Re: performance generally

[Emmanuel Lecharny <el...@apache.org>]

Andrew C. Oliver a écrit :

> As is probably obvious I'm working on TBird integration.  In comparing
> opends (https://opends.dev.java.net/) with the default configuration and
> example data the return results are much faster than ADS 1.0 (as noted
> 1.5 is not working ATM
> https://issues.apache.org/jira/browse/DIRSERVER-927).  Is it just a
> configuration, storage thing or index thing or a real differential?

I don't know. It will depend on the configuration (cache size, index, 
etc...). FYI, I'm running performance tests atm, and I must say that ADS 
is pretty fast on my laptop (Centrino 1,8 Hhz) : more than 4 000 random 
search requests per second.

At this point, you will have to post the configuration you are using, 
your schema (if any) and a sample of your data

>
> I followed the tutorial (as mentioned previously) then loaded the OpenDS
> example data into ApacheDS via the apacheds eclipse plugin thingy ldif
> export/import.  then I used a regexp
>
> For whatever reason a few entries didn't get copied to ApacheDS
> (probably I missed a limit to 1000 or something).  However that should
> have only affected the results in the other direction.

Which entries? Which error did you got?

>
> Searches on general things like A or B and specific things like Bonnar
> or "Bonnar Billoteau" show the same perceivable performance difference.

If no index are set on the associated attributes, and if your cache is 
small, and if you have a lot of entries, then performances will 
generally suck.

>  This bears out in both the GUI and Wireshark as to returned result
> timings.  ApacheDS handling thing connection better/faster but OpenDS
> handling the response in the sub-sub ms rounding error.

4000 req/s means is around 0,25 ms on ADS. Of course, this will depend 
on your CPU, memory, disk, network, etc.

Can you give us more information?

Thanks !