Shutdown events are typically written into points to indicate when the PI Server is taken off-line. This allows an end user to recognize the discontinuity of the data. Without shutdown events, the data looks like it has been interpolated over the time the PI Server was unavailable, which can be misleading.
Without buffering, it is clear when the PI Server stops running (shutdown or system crash), that every point which is collected on that system should receive a shutdown event. With buffering running, inserting shutdown events in points that are buffered gives an erroneous picture of the data. In fact, when the server comes back up, the buffered data is forwarded and there is no data loss. A shutdown event is irrelevant. The shutdown event can also cause buffered and subsequently forwarded data to be out of order, causing undue load on the PI Server and slowing down forwarding. For this reason, points that are buffered on an API node or home node should not have shutdown events written by the home node.