Address Assignment Question

David Miller dmiller at tiggee.com
Mon Jun 20 14:17:21 UTC 2011


On 6/20/2011 9:52 AM, Valdis.Kletnieks at vt.edu wrote:
> On Mon, 20 Jun 2011 09:26:30 EDT, Steve Richardson said:
>
>> *definitely* concerns me.  One thing they do say is that they need
>> several IPs per block to assign to their MTAs to handle such a large
>> amount of email (3 to 5 million per day).  Being primarily focused on
>> layers 1 through 4, I don't have an incredible amount of experience
>> with high volume email server configuration, so I have no idea if they
>> are feeding me a line of BS or not.
> It's BS.  5M a day is only about 60 per second, not at all a problem for a
> single IP address running properly configured SMTP software.
>
> For comparison, in the mid-90s, I was moving 1M RCPT TO's a day (and probably
> half that number of envelopes) on a Listserv host using Sendmail on an IBM
> RS6000-220 - a whole whopping 66MZ Power 604E processor and something like 64M
> of RAM (The same basic firepower as an old Apple 6600 Mac, if you remember
> them...)  Doing 10M messages a day on a single box is *easy* these days - the
> hardest part is getting a disk subsystem that survives all the fsync() beating
> most MTAs like to dish out....
>

Well... 10M messages per day on a single box today would be fine for 
hardware power, if most messages are accepted remotely on the first try, 
but not necessarily doable in the SMTP environment of today.  Mail 
servers that send a lot of email have to hold a lot higher percentage of 
messages in queue for longer today due to greylisting and other 
deferrals - particularly from freemail sites.

Your customer should only need X addresses per block for SMTP load 
sharing if they are going to have X number of physical servers.  If they 
are not going to have that many physical servers, then multiple 
addresses in the same block per server provides no additional throughput 
and could only be for block avoidance.  SMTP servers do most of their 
work managing mail queues - accepting new messages into queue, keeping 
track of messages in flight (those that failed and need to be retried), 
spoon feeding messages out to broken MTAs, etc... more IPs per box 
doesn't help this.

Someone who expects to be "blocked occasionally" would only need two (or 
a few...) address blocks.  Someone who expects to be "blocked all the 
time" would need *many* different discontiguous address blocks.

Are you getting spam complaints for their current blocks at an 
unreasonable (to you) rate?

Are they doing all the right things with SPF, DK/DKIM (not an invitation 
for a holy war on whether or not these are good or useful)?

If I put my tin foil hat on for a moment, I might suspect that your 
email marketer may be feeling the pinch of the economic downturn and 
might be considering implementing less scrupulous practices than they 
have followed in the past.  Even with my tin foil hat blocking out 
external voices... most internal voices agree that this sounds spammy.

-DMM





More information about the NANOG mailing list