In my case, it's filesystem. Is there a parameter that determines read-aheads? I'll look for HP-UX but if you know of a Linux or other OS parameter, it would make it easier for me to frame the question. As a result of this discussion I re-ran the "time dd" tests and found that the elbow is right before "bs=256k". Even "bs=255k" is faster (and as fast as 128, 32, anything less than 256). On Tue, 2006-12-19 at 13:03 -0800, Kevin Closson wrote: > Is this direct I/O or a filesystem? Could the filesystem > be firing off some uneccessary read-aheads when you do large > dfmbrc but not when you are nibbling with smaller multiblock request? > -- //www.freelists.org/webpage/oracle-l