si5: > >>May I suggest: you test the modified code and the unmodified code > >>and then try to explain why one is better than the other. > > >> Wietse > > Yes we have tested unmodified code with spirent(200,000 mails per 10 > minutes) and drops were very less.
That's 300/s, a performance level that Viktor reported for unmodified Postfix with a Dell server from 2003. https://groups.google.com/forum/?fromgroups=#!topic/mailing.postfix.users/pPcRJFJmdeA "One single Postfix instance has been clocked at ~300 message deliveries/second[8] across the Internet, running on commodity hardware (a vintage-2003 Dell 1850 system with battery-backed MegaRAID controller and two SCSI disks). This delivery rate is an order of magnitude below the "intrinsic" limit of 2500 message deliveries/second[8] that was achieved with the mail queue on a RAM disk while delivering to the "discard" transport (with a dual-core Opteron system in 2007)." > Ofcourse the unmodified code is better > but we modified it based on our requirements and now we are testing it too. > And it is showing significant mail drops. Once we are able make the drops > less we want to document the maximum load capacities of this modified > server. Thatswhy we are trying to find a document which has such information > so that we can do an analogous testing and documentation. There is no 'formula' to predict the behavior of a non-trivial program, especially not when the performance is determined by remote network performance, remore DNS server performance, and remote SMTP server performance. Meaningful numbers require meaningful measurements. BTW I would not consider a mail system as 'working' until all 'lost mail' instances can be explained. Your requirements may vary. Wietse
