On Fri, Aug 31, 2018 at 06:16:01PM +0000, Fazzina, Angelo wrote: > Hi, I was able to run a packet capture with tcpdump on the 3 load balanced > servers that handle massmail.uconn.edu during the users mail merge today. > It was looking like one email every 12 seconds
The main thing you're looking for is any substantial gaps between the client's TCP SYN packet and the server's 220 response, that might account for the low connection rate. If on the other hand you see the client disconnect at the completion of a message, and not reconnect for O(10s), then of course the message delivery will be low. Postfix can't accept mail faster than the client is willing to send it. That feature remains to be developed... > 3. This is instantaneous every time I test > > [root@mail4 ~]# telnet massmail.uconn.edu 587 That's encouraging. > 4. Servers are never that busy IMHO but I have 7 days of graphs that would > show that here [mail4,mail5,mail6] are the servers. > http://ssgunix.uits.uconn.edu/ssgunix.php Those are not the right metrics, what's relevant is the number of active TCP connections to port 587, and how close this number is to your process limit for that service (or default process limit). If you're getting close to capacity raise the process limit from the default (100) to 200 or even 500, assuming you have a server with fast disks (battery backed RAID conroller that batches and speeds up writes) and enough RAM and network capacity. If you never get close to 100, this is not your problem, but it may be prudent anyway. So far, it looks most likely that your user is throttled by his own client software. -- Viktor.
