NetNews Usenet Archive 1992 #18/NN_1992_18.iso/spool/comp/sys/3b1/3065

Internet Message Format | 1992-08-12 | 4.6 KB

Path: sparky!uunet!beartrk!ceilidh!hijo-2!dnichols
From: [e-mail] (Don Nichols (DoN.))
Newsgroups: comp.sys.3b1
Subject: Re: How to run 19,200 (or faster)
Message-ID: <[e-mail]>
Date: 13 Aug 92 00:29:47 GMT
References: <[e-mail]> <[e-mail]> <[e-mail]>
Organization: D and D Data, Vienna Virginia
Lines: 89

In article <[e-mail]> [e-mail] (Bruce Becker) writes:
>In article <[e-mail]> [e-mail] (Don Nichols (DoN.)) writes:
>|In article <[e-mail]> [e-mail] (Richard H. Gumpertz) writes:
>|>I will shortly be attcahing a 14.4 v.32bis modem to my 3B1.  How do I get a
>|>serial port to operate at 19,200 bps (or faster)?  I know this must be (or
>|>will be) a FAQ, but I couldn't find it listed.
>|
>|    The answer is "Not very well!"  You can get the serial port to run
>|at 19200, but the interrupt load (one per character transferred) is enough
>|to slow the machine down, and if you're also running the ethernet board, or
>|something else that uses a lot of cpu, you'll lose big time.  If the system
                                    ^^^

    I should have said "generates a lot of interrupts" here, since the
interrupts of the serial ports can always preempt user processes using the
cpu.  I don't have one, but I understand that the voice power board is
another big source of interrupts.

>|is lightly loaded, you'll simply re-try occasional packets, and not get the
>|throughput that you would like.

>    I am running a WorldBlazer and a Telebit
>    Plus at 19200 on thsi system, as well as
>    a direct link to another 3B1 at 9600bps.
>    This is a system with 2 disk drives that
>    are constantly active.

    The disk drives don't represent the kind of interrupt load that the
ethernet does, though the extra serial port with the direct link and the
other modem do, if they're both transferring at the same time.  The serial
cards generate an interrupt per character, while the disk i/o is typically
an interrupt per block of 512 or 1024 characters.  If the system is smart
enough, and the disk is sufficiently unfragmented, there is a possibility
for multiple sectors/transfer, though I don't expect that with the 7300/3b1,
since there isn't an extra cpu for disk control.

    I haven't tried what happens when running the floppy tape backup and
19200 uucp at the same time.  I suspect that it may be similarly
failure-prone.

    Also, I wonder whether the sites at the other end of the link are
actually managing to keep a 19200 link busy full time, or whether there is
some unused capacity which can be used for catchup time, so the data comes
out in spurts of 19200 at your end, but with gaps between spurts.

>
>    While throughput is not the greatest all
>    the time, the average for both modems is
>    about 750-800 chars/sec., and peak speed
>    for big batches is about 1350 char./sec.
>
>
>    If you run the ethernet driver the speed
>    slows to nearly nothing.

    Absolutely!  But if you drop the serial port speed to 9600, your throughput
improves dramatically *while* the ethernet is running.  I have had
connections fail while doing transfers at 19200 if the ethernet driver is
loaded, but never a failure (other than nearby lightning strikes taking out
the power at one end or the other) at 9600 *with* the ethernet.  (Of course,
I'm not doing a ping -f (flood) test to it.  When I do that from two Sun
systems (one 2/120 - 68010, and the other 3/140 - 68020) to one 3B1, (even
the one without the serial link to the modem), the kernel panics
spectacularly. :-)

>
>    A terminal server would be a pretty nice
>    solution but you need to be able to have
>    software which talks correctly to it for
>    outbound calls.


    The same for another proposed solution, which is perhaps more
practical given the budget that most users of 3B1s are working with.  That is
the possible use of the DOS-73 co-processor card as an intelligent I/O port,
to buffer transfers, and send completed packets instead of
interrupt-per-character once the flow is going.  We'd need a modified
version of uucp to work with it, but it offers possibilities.  I wonder how
the project is going.

    I want a source of breadboard cards for the 3B1, complete with the
unique connector. :-)
(Oh, yes, I also want kernel source at the same time. :-)

--
Donald Nichols (DoN.)     | Voice (Days): [phone] (Eves): [phone]
D&D Data                  |  Email: <[e-mail]>
I said it - no one else   |         <[e-mail]>
    --- Black Holes are where God is dividing by zero ---