at 150K users, the loads on my server aren't impressive, I'm guessing
Israeli users surf and chat more than write Emails, possibly because of
the software limitations (very few Right-to-left clients available, fewer
agree on the encoding of the characters)
My bosses are quite happy with an outgoing Qmail server, so now I want to
make all other functions work on Qmail (local delivery, virtual domains,
pop, ETRN users moving to AUTORUN etc.)
right now an ugly 8 meg password file with a 6 meg shadow sidekick are
pushed around the servers with scp. I'm going to move delivery and RADIUS
auth all to RDBMs... (anyone done this? It's really hard to find useful
info about this online... should I patch them all to lookup CDB files, or
lookup an SQL server maybe?)
the main question I'd like to pose to people, because getting sun machines
just for tests is too expensive an option here, has anyone compared the
speed advantage or loss when moving between the following setups:
1. current: sendmail delivers to a local in-house agent written in C (15k
tool) that tests for a vacation flag for a user, then delivers to a two
level hashed spool directory (/var/spool/mail/u/s/username) mounted from a
net appliance box after checking mail quota limits (not standard fs
quota). a second machine servers pop with qpopper.
2. wanted: qmail uses qmail-users or an external lookup (of CDB or some
SQL?) to deliver to a a single-UID hash of maildirs if within quota, while
checking for a vacation flag and executing if necessary. POP is served
from another machine using qmail-pop3d. no dialup users have a UID or an
entry in the /etc/passwd (YEAH!!!)
is qmail-pop3d up to such volumes? is the 2-order growth in number of
directories and files on the fileserver a speed damper? should I let qmail
deliver to the existing hash and keep Qualcomm's popper poppin'?
all sugestions and experianced tips are welcome, on-list or off it. TIA!
Ira.
(Oh yeah, and Russel, if you have a ready-made solution you can offer for
a fee, send me an offer!)
