Subject: Re: RFH: serial port benchmarking
To: Garrett D'Amore <>
From: Daniel Carosone <>
List: tech-kern
Date: 06/23/2006 21:21:53
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

On Tue, Jun 20, 2006 at 11:55:27AM -0700, Garrett D'Amore wrote:
> I'm looking for someone who can help benchmark UART performance to
> compare gdamore-uart and -current.  I suspect that the differences will
> be small, but I've been asked to provide some more concrete numbers,
> particularly of interest is i386 hardware, for which I do not have ready
> access.=20

I did some archeology, and dug up some old machines.  I have something
that I think is an 8Mb 33Mhz 486, and that still seems to power on and
POST get to the PRESS F1 TO CONTINUE prompt.  I'm not going to get
past this for a while, because it has a large-style AT DIN keyboard
plug, and I seem to have nothing functional to plug into it now.

It probably still has some ancient NetBSD, perhaps 0.9-ish era, on the
2x200Mb SCSI disks that seem to still both spin up.  I hope so, that
will make the problem of getting code onto it easier.

> (I also don't know how best to collect such performance data on
> i386 -- are there processor performance counters that can be used?)

My plan would be to set up some back-to-back serial transfers with a
modern machine, and crank up the serial speed until it starts to spew
about silo overflows.  If the current and branch kernels start failing
at different speeds, i'd call the effect noticable.  Suggestions for
other tests that I can perform without too much effort welcome.

> Also, if anyone has com(4) ports that are marginal at performance, it
> would be great to test them to verify that I haven't made things any
> worse. :-)

I found a whole lot of ISA serial cards, at least one with big chunky
83450's on, probably others.  It's not impossible that I have a HAYESP
around somewhere, too.


Content-Type: application/pgp-signature
Content-Disposition: inline

Version: GnuPG v1.4.3 (NetBSD)