When debugging through format ops tracing output, which is often overwhelmingly large, I commonly do something similar to grep <thread1> fulltrace > trace. thread1 and grep <thread2> fulltrace > trace.thread2 and then cat trace.thread* | sort > trace.mythreads

The problem is that the timestamps sort poorly because the microsecond portion is not zero-padded. It is a small change to add to the printf format string to print six microsecond digits and zero pad the output.