On Fri, Jun 01, 2001 at 02:38:04AM +0200, Karsten W. Rohrbach allegedly wrote:
> Mark Delany([EMAIL PROTECTED])@2001.05.31 22:32:26 +0000:
> > On Thu, May 31, 2001 at 11:13:56PM +0200, Roger Svenning allegedly wrote:
> > > Ok I see, so traffic shapers like altq and dummynet are made by people that
> > > don't understand the basics of tcp/ip ? :-)
> > > I didn't mean "blocked" literally, what I want is to make sure that smtp
> > > traffic, when qmail gets several thousand of mails dumped into it's queue,
> > > doesn't slow down http traffic too much, by putting some sort of a limit on
> > > qmail I want to avoid packetloss.
> >
> > We understand what you want. Do you understand that qmail has no
> > facility for doing this? The only way is to use a traffic shaper
> > external to qmail.
> qmail indirectly contains instrumentation for that. it is called remote
> concurreny.
No it doesn't.
> you might
> echo 2>/var/qmail/contro/concurrencyremote && svc -t /service/qmail
> which would limit the running qmail-remote processes to two which leads
> to less bandwidth consumption for outgoing mail.
Not necessarily and certainly not predicatably.
Tell me what happens with the following scenarioes:
Scenario one:
You have a concurrencyremote of 1
You have one email in the queue
That email is MXed to a yahoo.com address which has perhaps
a 1Gb or more of inbound connectivity
That email is 100MBytes in size
A qmail-remote is scheduled to delivery the email
Scenario two:
You have a concurrencyremote of 100
You have 100 emails in the queue
All emails are address to a dinky.connectivity.com. that
has perhaps 14.4Kb of inbound connectivity
Each email is 1MB in size
A qmail-remote is scheduled for every message in the queue
Question 1: What is the likely bandwidth consumption during delivery
for Scenario one?
Question 2: What is the likely bandwidth consumption during delivery
for Scenario two?
Bonus question: what part of qmail do you change to reduce the
bandwidth consumption for Scenario one?
Regards.
