This occurs typically cbus architectures such as the 7500. Your router receives a packet into a fast buffer on the incoming interface, and CEF transfers a pointer to the buffer to the outgoing interface. These fast output buffers are transmitted in strictly FIFO fashion.
However, CEF places a limit to the number of fast buffers that the output interface can hold on to. It does this for two reasons: because you don't want to run out of fast buffers, and because they are effectively bypassing any fancy queueing you may have configured. You can see the limit if you do show controllers cbus and look out for the word "txlimit". You can also tune it (at your peril) with tx-queue-limit.
The point is that once that limit is reached, the interface swaps the contents of the fast buffer out to normal DRAM, from whence the normal queueing machanisms start to cut in. It's a bit like a PC paging sections of memory out to the swap file.
"Output buffers swapped out" is really nothing to worry about. It's just how many buffers have been told to wait in the proper queue rather than being transmitted right away. It only gets to be a problem if they get dropped off the tail of the normal queue. I see you have dropped 3191 packets in that way, and that may be more of a problem.
If you want to read more about it, there's lots of goodstuff in the CiscoPress book on IOS Internal Architecture.
Kevin Dorrell
Luxembourg
