Great Circle Associates List-Managers
(March 2000)
 

Indexed By Date: [Previous] [Next] Indexed By Thread: [Previous] [Next]

Subject: VERPs versus batched delivery
From: Tim Pierce <twp @ rootsweb . com>
Date: Mon, 13 Mar 2000 04:05:58 -0500
To: list-managers @ greatcircle . com

John Levine commented recently about the common perception that
VERP-style delivery is a bad thing.  I have been dealing with this
perception here as well: I would desperately like to implement VERP
in SmartList, but the managers here are concerned about the effects
on our bandwidth.  However, none of us have any hard numbers to
back up our positions, so I did a little research tonight.

Unfortunately, the numbers I got appear to suggest that switching
to VERP deliveries would increase our bandwidth consumption by as
much as 40%.  Here's how I arrived at that number, for anyone who
wants to follow along and correct my assumptions.

          ------------------------------

I chose a thousand lists at random from one of our list servers.
(A more realistic study would have taken the volume of list traffic
into account.  That would actually increase the bandwidth consumption
even more, so my numbers may be considered conservative.)  For each
of those lists, I counted the total number of addresses subscribed
to the list, and the number of unique domains that those addresses
represent.

Because VERP delivery requires delivering a separate message body
for each recipient, the total number of addresses should be just
about the number of times we transmit a message body over the
network.  Batched delivery permits us to deliver a message body
just about once for each unique domain on the list, so I counted
the number of unique domains as approximately the number of times
we deliver a message body using traditional batch methods.

That gives me a way to estimate the bandwidth increase for any
given list: take the difference between the number of VERP deliveries
and the number of batch deliveries, and divide by the number of batch
deliveries to find the percentage by which that number would increase.

So I calculated this percentage for each of the thousand lists, then
added them all together and took the average percentage.  The result
was an average volume increase of 40%.

          ------------------------------

I dream about the benefits that we could get from VERP delivery --
reduced CPU utilization, increased server efficiency, less listowner
confusion, better word-of-mouth, and so on -- but a 40% bandwidth
increase means buying another T1 or more just to cover the extra
deliveries.  That's a pretty hard sell no matter what the bennies
are.

I'm interested in hearing from people who have migrated from a
sendmail-based mailing list platform to qmail, Postfix or some
other VERP-style delivery system, and have measured the bandwidth
deltas.  Did you in fact find a significant increase in volume,
and if so, how did you handle it?  (I am also interested in hearing
from people who can poke holes in my statistical methods, of course.)
 
-- 
Regards,
Tim Pierce
RootsWeb.com lead system admonsterator
and Chief Hacking Officer


Follow-Ups:
Indexed By Date Previous: Re: your mail lists on RemarQ
From: Rich Kulawiec <rsk @ gsp . org>
Next: Re: VERPs versus batched delivery
From: Russ Allbery <rra @ stanford . edu>
Indexed By Thread Previous: Re: your mail lists on RemarQ
From: Rich Kulawiec <rsk @ gsp . org>
Next: Re: VERPs versus batched delivery
From: Russ Allbery <rra @ stanford . edu>

Google
 
Search Internet Search www.greatcircle.com