Nice value worthless??? / Renice?

[Dave Martindale]

In article <1800006 at spdyne> root at spdyne.UUCP writes:
>
>	A few years ago, I was working on a VAX 11/730 running BSD 4.2.  When
>you would nice a process by say 2, you could tell the difference right away.
>It too considerably more or less time to complete.  A make niced to -20
>would zip along (and everyone else would stop).
>
>	Now I'm running a Compaq 386/20, with (sorta slow drives (40ms)),
>and only 3 Meg of ram.  Microport UNIX with DOS-MERGE. (Sys V)
>
>As far as I can tell, the nice value does absolutly nothing!

A 730 is *so* slow that you were probably waiting for the CPU most of
the time, and the nice value determines how much CPU you got, so it
determined responsiveness.

A timesharing 780 was usually like this too - a bit faster CPU, but
usually pretty fast disks and often several controllers, so you were
still mostly waiting for the CPU.  But on the Compaq, you've got a
processor that is faster relative to the disk.  So you're waiting for
the disk most of the time, the CPU runs whatever it can whenever it can
and still has lots of idle time, and nice just doesn't have any effect
- every process gets as much CPU as it can use.

What you really want is for niceness to control the priority of
requests in the disk queue.  But it doesn't.