Eric Thomas <ERIC@LEPICS>
Tue, 17 Oct 89 18:58:28 GMT
|
>CONNECT= in the response to CP Q T holds only two digits in the hours
>field. Thus if LISTSERV is connected more than 4 days the ratio is
>wrong.
Well, you're supposed to do an ACNT ALL every day. I could indeed get the
connection time from the "last IPL date/time" info that LISTSERV is
keeping, but that would assume that you log out LISTSERV every time you
reboot it, which I don't think is a good assumption.
>If I'm not mistaken EXPLODE doesn't consider Mail-Via=DIST2. The actual
>vs. calculated savings depend on the method of distribution.
This is absolutely true, but consider the amount of CPU time required if,
for each possible configuration, it would try out exactly how to spread
the people using the DIST2 algorithm. The way it is done, the distance
between the various recipients and the suggested peers is minimized,
which is about as accurate an estimate as one can get with a reasonable
amount of CPU time. One easy step in the right direction would be to
count all the recipients at backbone LISTSERV sites as being only 1
person, but somehow that's not going to be enough. Well it's relatively
easy to find out, for each recipient, the distance to the closest
backbone server, but that costs quite a lot of CPU time.
>Furthermore it doesn't suggest the movement (or connection for
>BESTPEERS) of peers.
Only BESTPEERS should suggest anything like that, and deciding on the
connection... well I don't know how other people do it, but personally I
pay more attention to the reliability of the links involved, amount of
horsepower available at the site and quality of postmaster support than
to the actual network load, since only 1 file is sent between the peers
:-) Would it be worth the work???
Eric
|
|
|