Summary: PPP and/or SL/IP between SparcStations via v.42bis modems

[Kelvin Don Nilsen]

Earlier in December, I had posted a query regarding the feasibility of
running PPP or SL/IP over a v.42bis modem between a Sun SLC at home and
Sun Sparcstation 1+ at work.  I had several specific questions, which I've
repeated below, along with the responses I've received so far.  Thanks
much for the information that's been sent.

I have decided not to use an SLC at home.  It appears that the SLC's
serial port is not able to effectively utilize the potential bandwidth of
a v.42bis modem.  Instead, I am planning to use an SBUS computer, with a
CoALM-S serial communications module purchased from AndataCo for about
$500.  This SBUS expansion board has a 10 Mhz RISC microprocessor to
provide intermediate buffering and control of 4 serial ports.  Each port
has a 16-byte FIFO.  Also, I've been assured by the salesperson that full
duplex RTS/CTS handshaking is supported.  I've ordered the board, but have
not yet received it.

Another stumbling block I've encountered now is that the only PPP
implementations I've been able to find are written for SunOS 4.0.3.  I'm
at SunOS 4.1.  Can anyone point me to a working implementation?

My original questions, with answers, follow:

1. Is it really true that the SLC (and SparcStation I+/II) do not support
    the newer full-duplex RTS/CTS protocol?

dd at mips.com (David DiGiacomo) tells me
|yes.

guy at auspex (Guy Harris) says:
|Yes.  NO Sun machine supports the full-duplex version, unless they've
|added it very recently.

a. If so, is this a hardware or software limitation?

dd at mips.com says:
|Hardware.

guy at auspex (Guy Harris) says:
|Well, it depends on what you consider "hardware limitation"s.
|
|The chips used on the Sun CPU serial ports and ALM-2 (Zilog/AMD Z8530),
|*and* the Signetics chip used on the ALM-1, do *NOT* include *ANY*
|"support" for full-duplex RTS/CTS, in the sense that they won't
|automatically somehow respond to full-duplex RTS/CTS signals.  They *do*
|include support for the half-duplex version, in that the transmitter won't
|transmit stuff if CTS isn't high, if RTS/CTS flow control is enabled (it's
|hardwired on for the Signetics chip; it can be turned on and off for the
|Z8530, which is what the CRTSCTS flag in "c_cflag" does in SunOS 4.x). 
|
|However, at least on the Z8530, they can supply CTS to the host and let
|the host wiggle RTS, so at least on the CPU serial ports the driver could,
|in principle, be modified to support full-duplex RTS/CTS.  I'm not sure
|whether the ALM-2 lets you do this.

b. I've seen mention of software patches available from, e.g.  Telebit to
support high-speed modem usage.  What do these patches fix?  this?

dd at mips.com says:
|Not this.

guy at auspex (Guy Harris) says:
|Dunno.  For UUCP, SunOS 4.x only requires that you have reasonably
|up-to-date serial port drivers (the 4.0 one had problems, patches exist,
|and the 4.0.3 and later ones may have the patches).  Earlier releases may
|have required patches to the UUCP software to let you request higher
|speeds.
|
|Other use of high-speed modems may have other patches.

3. What is the "architecture" of the serial port devices in the
Sparcstation family?  Does the CPU get an interrupt on every single byte
transmitted or recieved?  or do the ports have some sort of FIFO?  if so,
how large is the FIFO?

dd at mips.com writes:
|It's just a Z8530 -- refer to the data sheet for details.  It has a small
|FIFO, I think 3 bytes.

Wrote: Jim Hudgens <hudgens at sun13.scri.fsu.edu>
|I've played with SLIP between a PC and a SS1, and while I used xon/xoff
|exclusively, I can tell you for sure that the SS1 cannot function well
|above 19.2K when recieving.  I was measuring (as reported by vmstat) 30000
|interrupts a second at 38.4K, with about 70% system time, and no idle
|cycles.  It is better when sending.  

Wrote sharkey!ames!cheers!greg (Greg Onufer):
|The devices are Zilog Z8530's or reasonable facsimiles.  The have a 3 byte
|receiver FIFO (shift register + 2 bytes) and, yes, they cause an interrupt
|on every received character. 

guy at auspex (Guy Harris) says:
|Again, for the CPU serial ports:
|
|Yes, but it's a *very* high-priority interrupt, and the interrupt service
|routine does as little as possible.  For transmitting, it just grabs the
|next character from the current block being transmitted and hands it to
|the chip; for receiving, it grabs the character from the chip and stuffs
|it into a 256-byte circular buffer.
|
|When the block being transmitted is empty, or 16 characters show up in the
|receive buffer, a software interrupt is scheduled; that runs at a lower
|priority, and does the rest of the work.  The circular buffer is
|periodically drained, so that the system doesn't always wait for 16
|characters.
|
|The ALM-1 does, as I remember, do DMA transmission, although it interrupts
|the host on every character received.  I forget what the ALM-2 does.

4. PPP and SL/IP have some small degree of robustness.  Are they robust
enough to allow occasional loss of characters due to overflowing of
buffers?  (This is a subjective question, to be answered possibly by
someone who may have set one or two of these systems up in a configuration
that simply allows transmitting devices to ignore "XOFF" requests from
receivers)

Wrote sharkey!ames!cheers!greg (Greg Onufer):
|Yes and No, respectively.  SLIP does no link level framing/checksumming.
|PPP does do link level framing and computes frame checksums.  SLIP should
|not be used since PPP is now easier to come by than it used to be.  My
|Sun-2 seems to handle the Telebit TB2500 full-duplex hardware flow-control
|just fine and it uses the same UARTs as the current Suns (as well as the
|same device drivers (SunOS 4.0.3). 

5. Lastly, my main interest in this is to set up a computer at home that
provides the same X/NeWS user interface that I'm using at work and
provides access to all of the same data and executable files.  I do not
need a lot of processing capability at home.  If the SLC won't do it,
maybe I need to look for a cheap SBUS system like a used SparcStation I or
the new CompuAdd clone so that I can install a 3rd party "intelligent
port" board.  Any recommendations?

Wrote sharkey!ames!cheers!greg (Greg Onufer):
|A Sun-2/120 running 3.5 or 4.0.3 does this (not the NeWS part).  It will
|run an X11R4 server quite well.  This is what I do, anyways.  It's much
|cheaper (my Sun-2/120 with 80Mb disk, Mono monitor was $300--- your
|mileage will vary, of course).

Kelvin Nilsen/Dept. of Computer Science/Iowa State University/Ames, IA  50011 
 (515) 294-2259   kelvin at cs.iastate.edu  uunet!atanasoff!kelvin