The current log slot switching algorithm holds the slot lock while writing the content out to filesystem. That's necessary because a forced flush can come in, and it relies on an acquisition of the slot lock to indicate that any data in a slot has been written.

If we could make a change to the forced flush algorithm so that it no longer relies on the write happening inside the lock, we could remove a point of contention in our logging algorithm that is noticeable when the log is written on a slow filesystem.

This is related to WT-4058 and WT-4077. Those tickets have more complete description of a first attempt at this optimization, and the race condition it introduced.