Grouper Developers Forum

[Kathryn Huxtable <>]

Oh yeah, I'm still hoping to have a test version of this by Wednesday  
or Thursday of this week. -K

On Apr 7, 2008, at 1:49 PM, Kathryn Huxtable wrote:
> Whilst making my changes to ldappc to improve performance I found  
> two things it was doing that were even worse than I would have  
> imagined. One was in code that I was replacing anyway, but still...
>
> 1) It does linear searches through the members of a group to find  
> members to add or remove. This doesn't scale well at all and is  
> probably responsible for Cornell's (and my) failure to get  
> reasonable results when the group size exceeds 100,000 members. This  
> code wasn't going to be used in the new version anyway, but I had  
> imagined they were searching a hash table of some sort. Not so.  
> Unbelievable, IMNSHO.
>
> 2) Even if the comparison returns no changes, the code still does an  
> empty modify on the group. Again, unbelievable. Just putting a check  
> for an empty change array around the LDAP modification made it go  
> about three or four times faster through the groups on my test  
> system. (A 3.2GHZ dual-core Pentium4 wolfdale with 4GB of RAM and  
> 1TB of disk.)
>
> Anyway, you should expect some fairly large improvements even if the  
> rest of my ideas don't pan out, which they will because they did in  
> simple concept tests, and they're based on what Brown is doing,  
> which is fast.
>
> -K