On Tue, Jan 12, 1999 at 10:58:09AM -0700, Dax Kelson wrote:

> Firstly this is all academic, VGER doesn't run sendmail at all.  Its 
> mailing list performance is extraordinarily good considering the hardware 
> base and other load on the system.

Yep. Vger runs zmailer, which bogs down like all hell when it's queue
exceeds about 20k, especially when some addresses are hard to reach.
It also suffers from the bogus "optimization" of doing an extra dns
lookup to order it's queue.  This has been demonstrated to cost more
then it gains in real usage (or at least, I've proved it to myself).
 
> Second qmail is incredibly unsuitable for an arrangement like this.  If 
> you think otherwise you are not in the real world.  [Hint consider the 
> current bandwidth used by the current VGER config with distributed 
> exploders, now consider the fact you will multiply that bandwidth use by 
> several orders of magnitude if you use qmail].  

This strikes me as false, or at least incomplete.  Qmail has
additional capabilities that make remote list explosion pretty easy.
With serialmail and smtproutes a vger-like hub->exploder setup should
be doable with some work.  Probably less time would have gone into
developing the vger setup had qmail and its packages been around then.

Also, using a patched qmail-qmqpc (read the qmqp host from env or
file) and a qmqpd on the remote end list explosion could be made fast
and cheap - faster and cheaper then using smtp, anyway.  I think,
anyway.  I haven't had a chance to implement this.

> Qmail is not a seriously 
> scalable mailer - its a damn good single system performer, but it sucks 
> bandwidth for big lists and does not scale well.

Neither is zmailer.  In fact, zmailer is far less so, but they've
stretched it a long way.  In fact a lot farther then I'd have guessed
it could go. 

-Peter
 

Reply via email to