On Tue, 4 Oct 1994 15:54:45 +0100  Eric Thomas <[log in to unmask]>
had said:
>handful of servers only, so that's  no surprise. But the figures you give
>for the number of sites are very strange:
>
>>Values:  (number of sites counted)
>>    199312 : 274099.524 (21)
>>    199401 : 330140.125 (8)
>
>This suggests  that 13 1.8a sites  left the network with  VERS9312. Well,
>that has simply not happened so something must be wrong.
 
The reason why there are very few servers reported earlier are:
 
  * CTR data only became available after 1.8 was released, so not many
    servers had 1.8 running.
  * The earlier data was collected using an older version of LISTMON
    that used TELL to ask for this information, thus many attempts to
    obtain information failed; and it did not poll non-NJE Listservs
    at all. Now it uses MAIL to send its commands to LISTSERVs, and
    polls sites of verision 1.8 found in PEERS and INTPEERS NAMES files.
 
The bug with the graph originates from the fact that this month, the
CTR values were obtained twice, once one day before the end of the
month (because we needed the data beforehand), and again once after
the month was over. LISTMON caused multiple copies of this data to be
written to CTR DATA file, so the servers were counted twice.
 
I have since then cleaned up CTR DATA of this duplicate data, and have
put support in LISTMON server to UPDATE the CTR DATA file instead of
just appending the new information whenever a LISTSERV responds to
CTR polling.  I will post new figures today, after re-polling LISTSERVs
for their older statistical data, like 199312, 199401..etc.
From now on, even if a site is polled 50 times a month for the same data,
it will only be recorded once.
 
So, as a summary, this bug did not affect the numbers in "Listserv Daily
Usage" table (because duplicates are ignored for this chart),
and the "look" of the graph for the last month should be correct as well.
Since the server polled each site twice, counting them double, but then
divided the obtained figure by the number of sites (356?) reporting. Since
that chart is "per server" the plotted value should still be accurate.
 
The only part "very wrong" in the graph are the
  Values: (number of sites counted)
figures. They reported the counts for 199409 twice as high, and twice
as many sites.
 
(Btw, 186 are currently polled by the server, and each polling causes
MAILER@QUEENS to go into a crazy loop and spewing 20 MAILER rejection
messages.)
 
In any case, I'll be posting the new values today so we can compare
"before and after" values.
 
Thanks for looking into it, -turgut