"Erik E. Fair" writes:
> The Fast Filesystem has lots of knobs for tuning performance. I don't fully
> understand the effect of turning any one of those knobs in one direction or
> the other. How many people really do?
> 
> In one sense, disk technology has not changed much since Kirk did the work
> (we still have platters that spin with heads that seek), however, some
> things have changed (e.g. Fast SCSI, Fast & Wide SCSI, Ultra SCSI, IDE,
> EIDE, plus disks that spin at 4500, 5400, 7200, and higher RPM), so I
> wonder when the last time all the default parameters of FFS were revisited
> in light of those things about disk technology that have changed.

Things are even worse.

When the FFS was designed, disks had uniform numbers of sectors per
track.

Now, they no longer do, and the kernel is "deceived" into believing
otherwise.

It is unclear whether or not re-designing FFS to understand disks with
variable numbers of sectors per track would change performance, but
clearly things now designed to try to keep data on the same track fail
to do so.

> Further, given that full understanding of the knobs in FFS is limited, I
> wonder:  are there any tools that one can run on a raw disk (or on a
> filesystem) that can exercise the disk in some way, possibly analyze the
> filesystem if one is present, and then suggest potentially better FFS
> parameters, and project the likely performance increase?

There are a number of performance testing programs. Given these, one
can try twiddling knobs and "see what happens".

> Or are FFS's existing defaults already giving us 90% of the theoretical
> maximum performace for the general case (and thus such a diagnostic or
> exercise program wouldn't really be worth writing)?

I doubt it. As has been noted in the past, tuning often has dramatic
effects.

Perry