Skip site navigation (1)Skip section navigation (2)
Date:      Sun, 1 Jun 1997 23:47:59 -0700 (MST)
From:      Don Yuniskis <dgy@rtd.com>
To:        msmith@atrad.adelaide.edu.au (Michael Smith)
Cc:        freebsd-hackers@freefall.FreeBSD.org (FreeBSD hackers)
Subject:   Re: diskless hardware *design* suggestions
Message-ID:  <199706020647.XAA19653@seagull.rtd.com>
In-Reply-To: <199706010423.NAA11598@genesis.atrad.adelaide.edu.au> from "Michael Smith" at Jun 1, 97 01:53:49 pm

next in thread | previous in thread | raw e-mail | index | archive | help
It seems that Michael Smith said:
> Don Yuniskis stands accused of saying:
> > 
> > Haven't yet decided since that would be a daughterboard,
> > anyway.  BTW, the SC400 includes support for direct drive
> > of an LCD display, etc.  It *could* be kludged to drive
> > a CRT with a small effort...
> 
> ... so it has an onboard video controller of some sort?

Yes.  But, to be honest, I haven't looked at it in detail
as it's pretty minimalist (the SC400 is intended for use in
high end PDA's, etc. driving an LCD/pen interface).  Use
of the display controller requires you to give up the 32bit
databus in favor of 16...

> > > DMA is not common with NICs.  Shared memory (usually controlled by 
> > > the NIC) and programmed I/O are the norm.
> > 
> > Yes.  Those NIC's that support DMA tend to be bus-mastering
> > themselves -- hence my problem!
> 
> The Crystal parts do slave DMA, but that's generally too slow to be useful.

If they'll push 32 bits at a time and aren't limited by things like
the ISA "standard" DMA rates, that could be quite usable since it
would eliminate the need for a separate buffer memory, etc.

> > AMD's devices that are interesting all want to be bus masters.
> > Some of SMC's newer parts are somewhat appealing (10Base*
> > devices with integrated RAM, etc.).  Still no clear cut "winners",
> > though...  it's unfortunate that all the [34]86 MCU's have either
> > missing DRAM controllers or poor/nonexistent support for
> > bus mastering (obviously because they would have to drive the bus
> > back *into* the MCU core...)
> 
> TBH, unless you desperately need 100Mbps ethernet performance, a
> PIO-only solution using an 8390-family part (private SRAM) or even one
> using shared memory (remember that the 8390-family parts take most of
> the work out of arbitrating for the RAM, check the datasheets) will
> give you a very high-performance 10Mbps solution at low
> cost/complexity.

I'm suspecting that 1MB/s is probably less than ideal for some of the
applications I have in mind.  I'd like to nail down the network
interface "once and for all" and not have to redesign it later,
etc.  AMD's SuperNet2 parts look attractive but that's a sizeable
investment in silicon and driver development (and *definitely*
gobbles up too much real estate).  I was hoping a 10Base2
solution would be *very* appealing and keep me from digging
into some of the other options...  (hmmm... what's that?
"Design by laziness"??  :>)

> The "smaller" solutions like the SMC and Crystal parts are short on
> features (less packet RAM most significantly), and IMHO they're not
> suitable if performance is an issue.

Unfortunate but understandable.  I guess I'll start digging through
other network technologies and just plan on "biting the bullet"...

Thanx!
--don



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199706020647.XAA19653>