Developers of Sympa

[Olivier Salaün <address@concealed>]

Hi Elijah,

elijah a écrit :
> Olivier Salaün wrote
> > I've followed the track yourself and Sergiy explored, ie limiting the calls 
> > to List::get_lists() that browses the expl directory. I've just
> > commited my changes to the development branch and noticed another significant 
> > improvement of performances in the same conditions I mentionned  in 
> > myprevious email, ie 10 consecutive access to the service homepage on a 
> > server with 800 mailing lists. The initial process time was 7.2 seconds ; 
> > I've now reached 1.8 seconds. It seems to be resonnable performances, unless 
> > yourself and Sergiy still have insufficient performances with many K lists. 
> > If you could have a try
> > with the same profile procedure I used, see 
> > http://www.sympa.org/wiki/faq/performances ; I could thereby check that 
> > performances are now suitable.
> >     
> I believe we have a miscommunication. The current performance of wwsympa is 
> O(n) where n is the number of lists. Unfortunately, it will not work
> for us to have the time wwsympa takes to complete a request be proportional 
> to the number of lists, even if the execution time is very small for a 
> relatively high number of lists.
>
> Some people might find 1.8 seconds acceptable for a page request, but when 
> you start hosting 20K lists the execution time will grow unacceptably high so 
> long as the process is still O(n).
>   
Actually 1.8 seconds was the CPU time for 10 requests ; it also includes 
the time for the initialization of the wwsympa.fcgi process, done only once.
> Wwsympa should be able to support 100K lists, if the code is tweaked along 
> the lines suggested by Epsas and Sergiy.
>
> Riseup.net is a non-profit organization that hosts lists for other non-profit 
> organizations. We are working with Epsas to allow us to continue using sympa. 
> We are a small volunteer organization, and we do not have the resources to 
> maintain a forked version of sympa. For this reason, we hope and pray that 
> the performance improvements suggested by Epsas and Sergiy will be accepted 
> as part of the main sympa distribution.
>   
We too prefer that you run a "pure" version of sympa (I mean unpatched) 
and you're doing our best to integrate submitted patches, as long as 
they correspond to legitimate needs for the software. We had detailed 
feedback from Sergiy and Charles (alias Epsas) and we are now convinced 
that the proposed changes are the best solution to meet the needs of ML 
services with such big volumes. We'll have a closer look at the patch 
within the next weeks are we'll integrate the feature in the trunk of 
the project.

Thanks for your feedback
Thank you to Sergiy and Charles for their investment in the project.