Testing Tar (was Re: bad news for bsdtar..)

[Discussion moved to current_at_]

Bruce Evans wrote:
> 
> At least the -current version of tar skips reading the
> data when it is writing to /dev/null.

A-ha!  That explains a few of the odd timings I've seen.
I wonder why it does that?  (Other than to look good on
benchmarks, of course. ;-)

> I believe tar started being too smart under FreeBSD when it was
> imported into contrib.  Some of my benchmarks became invalid.

Care to share some of those benchmarks?  I'm especially
interested, of course, in the ones that did not become
invalid.  ;-)  Though knowing the ones that did might
be informative as well.

Generally, I'm looking for good ways to test the
performance of bsdtar.  As more people use it, I'm
getting more questions about performance, and it would
be nice to have some vaguely suggestive numbers to share.

Most of my tests so far are showing bsdtar to
be pretty comparable to gtar speedwise, but my
tests are mostly designed just to hammer on specific
subsystems.  I haven't made much effort to do broad
performance testing.

I suppose I should drag out my old tape drive, hook
it up, and experiment with that.  (Though I do get a
chuckle out of the idea of "performance testing"
using an old SCSI DDS DAT drive. ;-)

Tim