> > I am all set to migrate 350000 users to Qmail. I have created the
> > users with dummy passwords and while injecting their original password
> > noticed that it takes 3 hours to build the vpasswd.cdb file. After I
> > have built the .cdb file , if I change the password for even a single
> > user the .cdb file gets built again and worse it takes 3 hours to
> > complete the process. The result a password change comes into effect
> > only after 3 hours. Can someone substantiate if this is true
> > (hopefully it shouldn't be !!). Secondly if I kill the process then
> > the whole vpasswd file gets corrupt and I seem to have lost all the
> > passwords for the 350000 odd users. How do I solve this problem ??
I don't have a clear answer for you, but it is free :-) My recollection of
the code is that, for example, adding one user will cause the entire linear
vpasswd file to be copied (adding one line for the new user) followed by
pumping all the users into the .cdb file. I didn't think this was any big
deal because reads of the db are wayyyy more frequent than writes, but I
seem to have over simplified the problem.
350,000 users each with 100 reads per day of .cdb yields 35 million
reads/day at 20 ms per read or 700,000 wall clock seconds per day or about
8 wall clock seconds per second. (I've no idea what the numbers should be
but I think my assumptions are pessimistically high). That the good news.
If 1,000 users join, leave, or change their passwords, then you have about
3,000 wall clock hours per day or about 120 wall clock hours per hour
(again, I don't know what the numbers should be but I think 1,000 write
transactions is optimistically low).
I had been thinking that reads greatly outnumber writes, so no problem. But
my wild guess example has a read to write ratio of 35,000:1 yet the impact
is 1:15.
Where's my MySQL manual? I'm sure that the MySQL read time is no worse than
my pessimistic read example. And the writes? Well, it depends on how
vpopmail handles MySQL on writes. While your up, get those vpopmail sources
too, would ya?
--
Dennis Nichols
[EMAIL PROTECTED]
