Ioannis Tsouvalas put forth on 5/28/2010 4:09 AM: > My guess so far is to go and lower the link speed between the Shorewall, > Postfix, and maybe even SBS2008, and that's because similar problems having > been encountered where the MTU is bigger than 1500. Now, the only reason I > haven't done it so far, is that the MTU is already set at 1500, and I'm not > sure if dropping the link to a lower speed or replacing the vmxnet3 with > vmxnet2 that supports lower speed, would do any good.
The vmxnet 'NIC' is a virtual device, strictly a software driver. The vmxnet driver communicates with the ESX kernel at the speed of system memory, which on modern servers is over 10x faster than the 10 Gbe signaling rate. There is no such thing as "link speed" in this scenario as the interfaces are all software. Ethernet link speed, more correctly called a link pulse synchronization, is generated by hardware devices called PHYs. Link pulse is a hardware phenomenon. It doesn't exist in phantom software drivers, in this case the vmxnet drivers. Your issue is unrelated to the vmxnet "link speed" settings, unless there is a bug in the vmxnet driver code. If you send an email from an instance of Postfix running on a Linux guest to an instance of Exchange running on a Windows guest, both guests running on the same ESX physical machine, any communication between the two MTAs will occur via direct memory copy. The data will never be sent to the physical NIC in the server. The communication takes place through the ESX virtual ethernet switch, which again is strictly software. -- Stan
