There has been some confusion on this topic recently in the forums - of which I may have contributed to - so I am citing documentation this time.
dspfarm
DSP mapping occurs when DSP resources on one AIM or network module are available for processing of voice time-division multiplexing (TDM) streams on a different network module or on a voice/WAN interface card (VWIC).
dsp services dspfarm
Enables digital-signal-processor (DSP) farm services for a particular voice card/network-module.
So, in normal circumstances you would use the 'dsp services dspfarm' command to enable the DSP for MTP/Conference Bridge/Transcoding use. You would then give CUCM access to the DSPs through the dspfarm profile(s) and sccp ccm group commands. If you wanted to pool DSPs from the motherboard and/or network modules together you would use dspfarm. This is normally only done if you cannot get enough DSPs from one area or another to cover what you're trying to do.
It's also worth noting that each voice-card and the interfaces it supports (e.g. PVDM and EHWIC slots on the motherboard) can only handle a single external clocking source. You use the network-clock-select command to define what the source is for that clock. If you enable DSP pooling, then you have bound those additional PVDM/EHWIC slots on the network module into the same clocking domain. I am pointing this out because a common reason for separating E1/T1 circuits onto a network module is because they do not have the same clock source. If that's the case than DSP pooling would be a very bad idea.
