I found this answer at Professor Douglas Scott's FAQ page. He researches CMB and cosmology at the University of British Columbia.
How come we can tell what motion we have with respect to the CMB? Doesn't this mean there's an absolute frame of reference?
The theory of special relativity is based on the principle that there are no preferred reference frames. In other words, the whole of Einstein's theory rests on the assumption that physics works the same irrespective of what speed and direction you have. So the fact that there is a frame of reference in which there is no motion through the CMB would appear to violate special relativity!
However, the crucial assumption of Einstein's theory is not that there are no special frames, but that there are no special frames where the laws of physics are different. There clearly is a frame where the CMB is at rest, and so this is, in some sense, the rest frame of the Universe. But for doing any physics experiment, any other frame is as good as this one. So the only difference is that in the CMB rest frame you measure no velocity with respect to the CMB photons, but that does not imply any fundamental difference in the laws of physics.
“Where does it come from?” is also answered:
Where did the photons actually come from?
A very good question. We believe that the very early Universe was very hot and dense. At an early enough time it was so hot, ie there was so much energy around, that pairs of particles and anti-particles were continually being created and annihilated again. This annihilation makes pure energy, which means particles of light - photons. As the Universe expanded and the temperature fell the particles and anti-particles (quarks and the like) annihilated each other for the last time, and the energies were low enough that they couldn't be recreated again. For some reason (that still isn't well understood) the early Universe had about one part in a billion more particles than anti-particles. So when all the anti-particles had annihilated all the particles, that left about a billion photons for every particle of matter. And that's the way the Universe is today!
So the photons that we observe in the cosmic microwave background were created in the first minute or so of the history of the Universe. Subsequently they cooled along with the expansion of the Universe, and eventually they can be observed today with a temperature of about 2.73 Kelvin.
@starwed points out in the comments that there may be some confusion as to whether someone in the rest frame is stationary with respect to the photons in the rest frame. I found a couple more questions on Professor Scott's excellent email FAQ page to clarify the concept.
In your answer to the "How come we can tell what motion we have with respect to the CMB?" question, there is one more point that could be mentioned. In an expanding universe, two distant objects that are each at rest with respect to the CMB will typically be in motion relative to each other, right?
The expansion of the Universe is certainly an inconvenience when it comes to thinking of simple pictures of how things work cosmologically! Normally we get around this by imagining a set of observers who are all expanding from each other uniformly, i.e. they have no "peculiar motions", only the "Hubble expansion" (which is directly related to their distance apart). These observers then define an expanding reference frame. There are many different such frames, all moving with some constant speed relative to each other. But one of them can be picked out explicitly as the one with no CMB dipole pattern on the sky. And that's the absolute (expanding) rest frame!
Assumptions: From most points in the universe, one will measure a CMBR dipole. Thus, one would have to accelerate to attain a frame of reference "at rest" relative to the CMBR. Questions: Having attained that "rest frame", would one not have to accelerate constantly to stay at rest (to counter attraction of all the mass scattered around the universe)? [abridged]
I think the assumption is wrong, and therefore the question doesn't need to be asked.
The fact that there's a CMB dipole (one side of the sky hotter and the other side colder than the average) tells us that we are moving at a certain speed in a certain direction with respect to the "preferred" reference frame (i.e. the one in which there is no observed dipole). To get ourselves into this dipole-free frame we just have to move with a velocity which cancels out the dipole-producing velocity. There's no need to accelerate (accept the rapid acceleration you'd need to do to change velocity of course).
Our local motion (which makes us move relative to the "CMB frame" and hence gives us a dipole to observe) is caused by nearby clusters and superclusters of galaxies pulling us around. It's true that over cosmological timescales these objects are also moving. And so if we wanted to keep ourselves always in the dipole-free frame we'd have to make small adjustments to our velocity as we moved and got pulled around by different objects. But these changes would be on roughly billion year timescales. And so to get into the frame with no CMB dipole basically just requires the following 3 steps: (1) observe today's dipole; (2) move towards the coldest direction at just the right speed to cancel the dipole; and (3) maintain basically that same velocity forever.