My company has a new service that we are in Beta with that provides mass 1 to 1 email capability. On Dec 23rd our beta customer sent Holiday messages to 4,000 of his customers, the queue backed up and it took 2 hours for the email to be sent. These were short text only emails without attachments. Any thoughts on improving performance and would moving to Exchange 2007 with 8gb of memory help. For this service to be a viable business model we need to be able to send out 500,000 email as described above in 8 hours. Thanks

It shouldn't have taken that long. Did you look at any of the messages in the queue to see if there had already been an attempt to deliver them?Outlook Web Access For PDA , OWA For WAP www.owa-pda.com email a@t leederbyshire d.0.t c.0.m

Yes and they were delivered on the first attempt.

So you are fairly certain that the messages in the queue had not been held up externally for some reason?Did you configure a connections limit on your SMTP Virtual Server?Outlook Web Access For PDA , OWA For WAP www.owa-pda.com email a@t leederbyshire d.0.t c.0.m

Yes we have unlimited SMTP connections and we verified that the messages were not being held up externally. Also, is it a single processor server with 4gb of memory.

If you run the Task Manager when you do a mailshot, does it look like either CPU or memory usage is running at or near 100% for a long time? Or is this not something you can easily check till the next attempt (I realise you may not want to send out another load of mail and risk getting blocked)? 4GB should be enough for E2003, but Task Manager or Perfmon are the only ways to tell if the server is running short.Outlook Web Access For PDA , OWA For WAP www.owa-pda.com email a@t leederbyshire d.0.t c.0.m

We didn't check either and you are correct we cant test until May when the next big mass mailshot goes out. I guess the question is can someone tell me they have a config that can handle 500,000 outbound emails in an hour. Thanks Ed

I don't know if you already tried it, but there is a tool named loadsim that is supposed to help in these situations:http://www.msexchange.org/tutorials/Simulating-Stress-Exchange-2003-LoadSim.htmlbut you may find that something external to the server is having an impact. E.g. your firewall may be configured to throttle SMTP connections. Or your ISP may even be doing it, to prevent people from unknowingly doing what you are trying to do.Outlook Web Access For PDA , OWA For WAP http://www.owa-pda.com email a@t leederbyshire d.0.t c.0.m

Lee: Thanks for those tips. We will give them a try. Hopefully someone will reply that has something similar working already. Ed

I'd turn on protocol logging, and see if you're being throttled.

Thanks for the tip on Protocol Logging....we will turn it on. Ed

On Wed, 13-Jan-10 15:42:34 GMT, Lee Derbyshire [MVP] wrote:>If you run the Task Manager when you do a mailshot, does it look like either CPU or memory usage is running at or near 100% for a long time? Using PerfMon would be a better choice, I think.Put these two together and you'll be able to get a pretty good readingof what's going on on the machine before you run into another problem:http://blogs.technet.com/mikelag/archive/2009/02/02/updated-exchange-2003-perfwiz.aspxhttp://www.codeplex.com/PALHaving the data collection running while your machine is doing theheavy lifting will give you a much better idea of where the bottleneckis (if you haven't already found it before then).>Or is this not something you can easily check till the next attempt (I realise you may not want to send out another load of mail and risk getting blocked)? 4GB should be enough for E2003, but Task Manager or Perfmon are the only ways to tell if the server is running short.I agree. The machine should have blown through those messages prettyquickly. Of course, all we know is that the machine has a single CPU(not a good idea -- Exchange reacts well to being able to perromconcurrent activities) and 4GB of memory. No mention is made ofconnections speeds, latency, or bandwidth. No information waspresented about disk perfromance (seconds/IO, IO/second, number odspindles, etc.). If this is all being done on a single machine, wasthe data sent to it using SMTP or from a MAPI client (which wouldinvolve the mailbox database).Exchange 2003 was a big improvement over earlier versions when it cameto the number of disk I/Os that had to be executed as a SMTP emailpassed through the system, but there was still a lot of it going on.---Rich MatheisenMCSE+I, Exchange MVP --- Rich Matheisen MCSE+I, Exchange MVP

During the last Holiday send, the CPU did not appear to be very High. We should be able to highly monitor this on the next send or we can setup a test environment to do so. We recently upgraded to new firewall and will be giving that a run down as well. As mentioned above, our Exchange does run on a Single CPU server and has 4GB. We will double check our connection speeds to see if that is slowing us down. Would going to Multi-Processors greatly enhance Exchange's performance? The Data was sent using SMTP. Your help is greatly appreciated. Ed

On Wed, 13-Jan-10 20:52:26 GMT, meaEd wrote:>During the last Holiday send, the CPU did not appear to be very High. We should be able to highly monitor this on the next send or we can setup a test environment to do so. We recently upgraded to new firewall and will be giving that a run down as well. As mentioned above, our Exchange does run on a Single CPU server and has 4GB. We will double check our connection speeds to see if that is slowing us down. Would going to Multi-Processors greatly enhance Exchange's performance? The Data was sent using SMTP. Your help is greatly appreciated. Ed Anyone can guess at what your bottleneck is, but the only way toreally know is to measure.Does Exchange do better with multiple CPUs? Sure -- if the workload isenough to cause either delayed I/O operations or queues of operationswaiting for the CPU. But if your machine is cruising along and onlybogs down under load (and 4,000 messages isn't what most of us wouldconsider to be much of a load) the only way to know what's going on isto burden the machine and measure what's going on and then correctwhatever is out-of-spec. But, as with all perfromance tuning, thatjust eliminates that bottlenecek -- and there's always another on justwaiting beyond that one.I guess the best way to say this is that you don't size a server foran average workload, you size it for the peak workload.---Rich MatheisenMCSE+I, Exchange MVP --- Rich Matheisen MCSE+I, Exchange MVP