On Sun, 29 Aug 1993 23:25:12 +0300 Matti Aarnio said:
>> I run two large (~500 international recipients) lists on a DECstation
>> 5000/240 running Ultrix 4.3. The lists are managed by Majordomo, but
> Like most of us do :) -- Really large lists do need alternate methods
>for routing, as you have seen.
>> Everything works fine, but I'm wondering if there's any way to improve
>> the turnaround time. The delay, according to syslog, ranges up to 3
>> hours [...]
> There is a system which handles several million (aggregate) messages
>a day on long lists, but its trick is usage of static routing tables.
>(Yes, it is the "Revised LISTSERV") I don't believe there is a way to
>achieve equivalent situation in the current DNS managed universe. When
>everybody gets their act together at running the DNS, and understands that
>every mail-receiving system MUST have MXes, and they have their domain
>zones replicated (with authenticity from the upper level!) so that there
>would always be data available.
>To summarize, your alternatives are:
Multi-tasking on a single machine may help, but it has its limits.
*IF* the original MLM is part of a network of MLMs, then there may
be another alternative.
Split the list of subscribers into sections, and let MLMs on OTHER
HARDWARE take care of distributing the item to their subsection of
the subscriber list. This could be done by either distributing the
item with a subsection of the subscriber list, or having each
participating MLM maintain its own section of the subscriber-list.
For either method, each separate MLM may need to advertise its own
"throughput factor", ie. how many outgoing items it can handle in
a given unit. This would be useful in determining over how many
(and with which) MLMs to split a subscriber-list.
Of course there's the possible problem of sites not wishing their
own resources to be used in this manner by other sites.
But anyone running or having run a Revised LISTSERV in the BITnet
world knows that such cooperation can and has been achieved, to the
ultimate benefit of large numbers of people & less resource wasting
in the overall sense.
How do the various MLMs handle an item destined for more than one
subscriber at the same site (ie. only the local part differs)?
The Internet is basicly a connection-oriented network. But this
setting-up of a connection takes real time (and not small fractions
of a second either!).
There's a bit of wry humor in that a solution for a throughput
problem in an originally rather slow connection-less daisychained
network might be useable as solution to essentially the same kind
of problem in the TCP/IP / DNS world. :-)
Not too sure, but RFC-1429, the "Listserv Distribute Protocol" by
Eric Thomas might shed some more light on the whole subject.
Actually, I'm rather shocked that this imho rather obvious solution
hasn't yet been suggested and implemented in the Internet world.
Will someone please prove me wrong in some way?!?
To cooperate or not to cooperate, zat ies ze qvestionne... ;-)