I think using locking on the logfile will be a bad idea, as it will/could create logfile contention, as the processes will have to wait to commit a line to the log. In theory (that is, tye said so) appending a line to the logfile is atomic and we shouldn't need any locking anyway. Keeping the FH open will be bad mojo though - maybe we should do the ugly dance of seek-to-end, syswrite. But I would like to see arguments/statistics that the opening/closing of files is a major slowdown factor over the locking. Since the multiple lines get spewed into the logfile directly anyway, I don't expect a major speedup, but maybe we could switch the whole stuff to syslogd or some other log aggregator mechanism that is easily available on pairs servers already...