>Inbound will be fine once I dedicate the current outbound box to inbound >only. >It's the outbound / ETRN that takes all of my resources.
outbound from the mailbox server is typically less than 20% of MX inbound to the mailbox server. so if two boxes can handle the inbound for your increased traffic, I don't see why two other boxes won't loaf handling your outbound >My thought was 2 TigerDirect type servers dedicated for this task. >Just not sure of the config... They will build to spec. I have partner who bought a Tiger Direct Vision box for one his clients as an outbound gateway with me doing the admin. I spec'd a Promise TX2000 for the ATA133 support of ATA133 drives. ATA was stuck 33, until we figured out Vision had reversed the ATA133 cables end to end. Now we are at ATA133. >My thought was to have a second box to be able to balance the increased >load. if you have two boxes for load balancing, and one box goes down, then the remaining box has to be powerful enough to keep up with the entire load. ie, the box was carrying 50%+ of the load, now has to carry 100% if you had 3 boxes, and one goes down, then the 2 remaining boxes increase their traffic from 33% to 50% of total. the point: 2 boxes is not a good load-balancing config unless each box it very powerful. iow, if one box is not enough and you want to spread the load, you must had 2 boxes (3 total, not 1 (2 total). ie, you always need n+1 boxes, so when the "1" goes down the "n" can carry the full load. >Processor of the day ofcource, what I need are the drive configs >Can you define for me the config: > Controller suggestion / RAID type If you stay with FreeBSD, then check the LINT file for supported SCSI controllers that do caching. These boards are typically $500 to $600. A lot of these controllers are 64-bit, 66 MHz PCI boards which would be happiest on an equivalent mobo. This is for the mailbox server, not for the smtp gateways RAID type is 1 for SMTP gateway. for the mailbox servers, you will probably need striping and well a mirroring to get enough capacity for the mailboxes. >and what channels are connected to >which drives. I see no value in internal $$redundancy of MX boxes. The boxes are redundant at the box level. the MX algorithm and DNS gives you inbound redundancy. For outbound redundancy, Imail's "send all mail through gateway" doesn't do MX lookups, it does A lookups, so the MX algorithm doesn't help on outbound. If you want dynamic load balancing for the outbound (or the inbound), then I have developed dynamic DNS scheme adapted to postfix (ie, it's "postfix aware". Imail would send to the domain: mxout.snipmail.net. 5m ip.ad.re.ss and my DDNS would change the ip.ad.re.ss to that of the gateway box where postfix is least loaded. This includes NOT using the ip.ad.re.ss of a postfix box that has died. Like a Radware Linkproof box, but a lot cheaper. :) Len
