That paper was the first thing I thought of when I read this, too. It really is a fascinating solution to this problem. Any ideas why external synchrony didn't

That paper was the first thing I thought of when I read this, too. It really is a fascinating solution to this problem. Any ideas why external synchrony didn't catch on? I suppose one could argue that the implementation would add a decent amount of conceptual overhead to the kernel, for comparatively little reward (application authors can fsync now, etc.).