Idea for improvement (Doesn't seem to be doing this already) is to
checkpoint the log file every n MB.
This would allow one to recover from a network dropout without
starting over from the beginning.
(if for example you are writing to a file on another server).