What is the difference between bonding and load sharing?  

Do I need a BODi at only one end or at both ends?

1) When you have a BODi at only one end of a connection, you can do load sharing over multiple GSM connections. Let me explain this in more detail. For example, consider having two GSM connections, which in fact, could be different carriers. You start a video session and it will be transported over the first GSM connection for the entire session, until it is finished. If you start a second video session while the first continues, it will go to the second GSM connection for transport. Therefore the load is shared over the two GSM links, but each session is one, and only one, GSM link.

2) On the other hand, Bonding requires a BODi at each end. (In fact, true bonding of multiple streams needs to have some type of device at each end, such as in MLPPP.) In the bonding scenario with a single video session, the packets (or frames) are distributed in round-robin fashion to the multiple bonded GSM links. In other words for an example with three GSM links, packet #1 goes to GSM link #1, packet #2 goes to GSM link #2, packet #3 goes to GSM link #3, and then it repeats by sending packet #4 to GSM link #1, etc.
In this scenario, the single video session would use all the GSM links (that make up one bonded point-to-point connection) to carry the video traffic.