[Raw Msg Headers][Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: parallel smtp sessions to same target ost

> Well, to make it easy to understand, an example:
> *.customer.nl		IN	MX 10	relay.customer.nl
> *.customer.nl		IN	MX 50	smtp.NL.net
> 	When customer.nl relay is down a lot of mail is spooled at
> smtp.NL.net If there are a dozen of huge mails there, all the small
> ones queued afterwards would get pretty delayed. Even if you reorder
> your queue file by size, if some small mail received while you are
> delivering your huge one they will get stuck there for a while. What
> I need it to have parallel smtp sessions open.

	What you really want is to have multiple parallel threads
	to same host with priorization of selected jobs fed to them.

	We could solve this by having a way to build separate threads
	based not only on channel+host pair, but also to some other
	criteria -- like size-bins ?  How you would configure them ?

	Things are simplified, if the selection criteria is same for
	all recipients on same channel+host pairs within one message.
	I mean that it should place all recipients at same host into
	same thread.  If it doesn't, well, scheduler and transport-
	agent protocols need major overhaul job...

	For that matter, is there any known way to do it with UUCP ?
	Multiple parallel threads feeding stuff ?

	I do know that NJE networks (like BITNET/EARN) had NJE sub-stream
	priorization schemes based on varying things, sender/recipients,
	spool class, file sizes, file ages, ...
	Way back EARN used to feed 7 parallel NJE substreams with 4 of
	them feeding small messages (under 500 punch cards, or so),
	2 did feed mid-size ones (under 50 000 cards), and one feeding
	huge ones...  Those were 9600 bps BISYNC links, btw.  Latter
	they were upgraded to 64000 bps SNA/VTAM pipes, but still the
	same basic protocol transported the files.
	( If you want to see weird code, see  FUNET-NJE running on UNIXes,
	  and speaking NJE against Big Blue iron -- it is even better at
	  talking to MVS machines, than VM/SP systems were. )

	I don't have time to look at that code before week 6, after
	I have returned from RIPE meeting at Amsterdam.
	(If any of you are around there, we can meet.  I will be flying
	 in on 27th.  RIPE days are 28-30th.  I am going in for the
	 SPAM-BoF and IPv6..  http://www.ripe.net/ )

> 							alexis

/Matti Aarnio <mea@nic.funet.fi>