[Top] [All Lists]

Re: Deliver-qmails-ondemand

2003-10-28 12:31:04
On Wed, 29 Oct 2003 00:14:22 +0530, Madan Ganesh Velayudham said:

      I may wrong here. Please correct me if I'm.
      When a server comes up, a separate thread would take care
      sending #INVITE message to its neighbours and Parallely
      it can take care other activities.

Yes, you *could* do that.  The problem is that if you have been down for more
than a few hours, having a separate thread that sends ETRN to all your peers
will result in your MTA receiving all the several hour's backlog in a few 

It's not exactly clear how to do proper rate-limiting in this case, or even if 
should be attempted at all.  You need a very good predictive function for system
load that knows where the "knee" in the performance curve is.  The problem
is that said function has to be quite reactive (reflecting the past few seconds
of performance and an accurate prediction of the next few) but at the same time
it isn't prone to false positives (throttling all inbound mail because of a very
short term issue that will be cleared in 5 seconds or so).

Somebody needs to take all this sort of stuff and publish an Informational RFC
on Operational Considerations....  Any takers?

Attachment: pgpu83NPTJo0f.pgp
Description: PGP signature

<Prev in Thread] Current Thread [Next in Thread>