The mailing list for listmasters using Sympa

[David Verdin <address@concealed>]

Hi,

Liam Kirsher a écrit :
> Maybe this was a bit vague, but I would like to know a bit more about
> how Sympa handles large lists!
>
> ## Maximum number of recipients per call to Sendmail
> nrcpt   25
> ## Max. number of different domains per call to Sendmail
> avg     10
> ## Max. number of Sendmail processes (launched by Sympa) running
> simultaneously
> ## Proposed value is quite low, you can rise it up to 100, 200 or even
> 300 with powerfull systems.
> maxsmtp 100
>
> What kind of throughput should I expect with that hardware and those
> settings?  Is there some documentation on tweaking these values for best
> performance?
>   
Not in the Sympa site. I'm afraid.
Do you think it would be useful to have a page where people could 
indicate the best values they found out for their installation? It could 
contain a simple array with fields such as :
- server hardware
- Sympa version
- MTA
- nrcpt value
- avg value
- maxsmtp value
- distribution_mode value
- number of lists in the server
- max number of subscribers for one list

This page could be hosted in our wiki, so that it is easy to keep it 
up-to-date.

Note that avg determines actually the max number of addresses for each 
domain in a call to the MTA. This is intended to prevent being 
considered a spammer.
nrcpt relevant value depends on the MTA you're using. It won't have a 
value over 40 with sendmail, because sendmail will start working as soon 
as it is called. It can be much larger for postfix which uses a spool.
As for the max number of processes: it all depends on your system's 
preformaces. If you reach the max number of processes often and for a 
long time, you can try increasing it and see if your machine handles it 
well. Please note that tweaking the lists' priorities can help, too. 
Indeed, when Sympa checks a spool for messages to distribute, it will 
first distribute the messages for lists with the lowest value for 
parameter "priority".

As a side note, there is this faq about evaluating performances in WWSympa :
http://www.sympa.org/wiki/faq/performances

I can write a page about the various parameters and mechanisms involved 
in mail distribution optimization in Sympa.
> Also, how exactly does Sympa deal with that many messages?
> If, for some
> reason, it were interrupted would it be able to continue where it left off?
>   
Not yet. This is something we are working on, but the exact means of 
doing it is still debated.
For now, Sympa won't stop distributing until the message has been sent 
to all the message recipients.

Near future works will be to allow Sympa to restart working where it 
left when interrupted and also, probably, to allow nightly distribution 
of messages for a given list, which involves introducing fixed time 
(instead of fixed time interval) tasks in the task manager.

Regards,
> Thanks in advance for shedding some light on this for me!
>
> Liam
>
> Liam Kirsher wrote:
>   
> > Hi --
> >
> > I wasn't able to find any performance statistics, so maybe someone can
> > give me a general idea...
> >
> > I'm running Sympa 5.3.b4 on an older RedHat 8.0 system.  A small number
> > of relatively small (<2000 subscriber) lists.  The server has two 650MHz
> > CPU's and 899M memory.  (I have a new server I'm installing which is
> > more robust, but that won't happen for another month.)
> >
> > Currently, the server is not very busy.  I have a celebrity client with
> > a mailing list of about 80,000 addresses. 
> >
> > Will that server be able to handle a list of 80,000 addresses?
> >
> > I'm a little unclear on how the emails are delivered to the MTA.  They
> > are batched, so that the MTA (postfix, in this case) doesn't try to
> > deliver them all at once, right?
> >
> > I'm also running dkim, which requires some resources, as well.  (And on
> > the receiving end, Amavisd, SpamAssassin, ClamAV.)
> >
> >
> > Another question -- what are commercial services that provide this kind
> > of service?  I want to charge a fair market price for running this for them.
> >
> > Best,
> > Liam
> >
> >
> >   
> >     
>
>   

--
David Verdin
Comité réseau des universités