Re: Virtual users pop3d suggestions

[Jeroen Geilman]

On 09/22/2010 02:22 AM, Noel Butler wrote: On Tue, 2010-09-21 at 16:47 +1000, Nick Edwards wrote: On Tue, Sep 14, 2010 at 4:53 AM, Seth Mattinen <se...@rollernet.us> wrote: Thanks, we have over the weekend ran two testbeds at full thrashing with in house written scripts, the timings show after 57 hours of constant stress tests with identical copies of various messages pop'd by both using 1000 parallel accesses, for pop3 courier is no faster than dovecot, we are sure if it was imap it would be a different story, but we have no use, since sqwebmail uses pop3, we can eliminate imap completely, the decision to our problem is simple now, after this test, we see no reason to continue to use dovecot in its current state with its inherit risks when courier has none of them, the move to courier is now justified. Thanks to all who offered alternative suggestions. I too am considering courier due to dovecots pitfalls, we used it early part of the millennium with qmail ourselves, it was good, despite it being very robust, I never really liked sqwebmail    but... since an unmeasurable percentage of users use webmail, its neither here nor there as far as I'm concerned. I'd be interested in seeing the results of your tests if possible, off-list is fine if you want, it might help sway my decision, I like dovecot, but a flaw that can be worked around but wont be worked around is a flaw none the less, it might be nit-picking, but it is there, it's always going to reindex its UID files in pop3 as well as imap, I pride myself in having a faultless system, even though there is little risk with pop3, it is a risk none the less, a risk that does not exist using other software. As already explained earlier, if there is a problem with high loads on a docevot (pop or imap) + NFS + webmail system, don't blame dovecot. IIRC NFS has ALWAYS had issues with high concurrency and mailbox sharing. It would be very instructive to see empirical proof of this issue (proof that it lies with dovecot deliver), since I am sure dovecot's developers would appreciate such feedback. -- J.