Some test-results of SMTP PIPELINING

[Matti Aarnio <mea@nic.funet.fi>]

Hello,

	Here are some notes about ZMailer at severe overload, (also
	known as  vger.rutgers.edu :-) ) of what I have talked with
	David Miller at Rutgers, therefore these look a bit ephemeral,
	but bear with me.

...
	The scheduler was offline for a while after I had practically
	forgotten that I closed it for config change (individual log-
	files for smtpx processes -- very informative logs :)
	/var/log/mail/smtpx.* )

	It tells typically following story about message transfers:

2505 bytes, 300 in header, 1 recipients, 0 secs for envelope, 1 secs for body xfer
1723 bytes, 1003 in header, 70 recipients, 2 secs for envelope, 1 secs for body xfer
361 bytes, 294 in header, 1 recipients, 1 secs for envelope, 0 secs for body xfer
1696 bytes, 1005 in header, 70 recipients, 15 secs for envelope, 1 secs for body xfer

	I just wanted to see, how fast the PIPELINE stuff is :-)
	(The log has a lot more stuff, those lines are just extracts.)

bash# egrep envelope smtpx.nic.funet.fi| awk '{rsum += $7; tsum+=(($9)+($11));}
END{printf "NR %s rsum %s tsum %s\n",NR,rsum,tsum}'
NR 926 rsum 65930 tsum 2188

	That is, 926 messages with 66k recipients in 2200 seconds (approx)

	For comparison a target without PIPELINING :-)

bash# egrep envelope smtpx.mailboxes.nether.net| awk '{rsum += $7; tsum+=(($9)+
($11));}END{printf "NR %s rsum %s tsum %s\n",NR,rsum,tsum}'
NR 690 rsum 31792 tsum 52973

	That is, 690 messages of 32k recipients in 53000 seconds -- apalling..
...

	I am trying to see, of I can make ZMailer Router also smaller;
	ZM has now a heap-based pseudo-LISP (data objects) in it, I am
	writing a full GC-based memory handling for it.
	(Actually it is rather simple to do, like LISP always, just needs
	 keeping track of variables pointing to objects..  Now maybe I
	 can shrink the bloated memory of router..)
...
	/Matti Aarnio