🚀 go-pugleaf

bobk@dogear.spk.wa.us (Bob Kirkpatrick) writes:

> geoff@zswamp.UUCP (Geoffrey Welsh) writes:
> 
> > sjsobol@tiny.com (Steven J. Sobol) writes:
> > 
> > > Hi. I'm running Waffle 1.65 and I am getting 800-1000 cps with uucico, bu
> > > I think I can do better. I'm using a Boca 14.4 internal fax modem with
> > > a 16450 UART. Any suggestions?
> > 
> >    Yeah: disable error correction.  Standard implementations of UUCP 'g' 
> > protocol has a limited window (usually 3 packets at 64 bytes, sometimes 7 
> > packets at 64 bytes), and the increased latency caused by error correction 
> > holding back data until an entire packet is received and CRC-verified can 
> > result in poor performance.
> 
> Are you sure? I run a Supra with the whole package turned on, and get ~1600
> cps...
> [...four lines down...]
>
> I run 7 windows and 1024 packets. The 3/64 is almost 1/3 slower.

   That (your 1024 byte packet size) is exactly why disabling error correction 
is unnecessary in your situation.  1/3 slower than 1600 CPS is just over 1000 
CPS, the rate that Steven complained about... and which may be improved by 
reducing the latency using one of the approaches I suggested.

> >    Less drastic: configure MNP4 to use small packets only.  This will speed
> > the release of UUCP packets in a long data stream.
> 
> I run 7 windows and 1024 packets. The 3/64 is almost 1/3 slower. However, if
> one has a particularly noisy line, then the smaller setup may be better be-
> cuase of the fewer bytes needed to be retransmitted. But then, if you have
> a dirty line, you aren't going to get optimum throughput anyway. ;-/

   Don't confuse MNP packet size with UUCP packet size.  By "configure MNP4 to 
use small packets only", I meant that Steven should, if possible, disable 
adaptive packet assembly in the modem, which can vary the MNP packet size from 
64 bytes (the 'standard' without MNP4 APA) up to 256 bytes.  Doing this 
increases throughput when using streaming protocols like ZMODEM, but it will 
'absorb' an entire 3-packet UUCP-g window into one large MNP packet, transfer 
it (very efficiently), and release it in one burst.  It's better to lose a 
little bit of efficiency and force MNP to release the UUCP packets in 64-byte 
chunks, so that there is less of a delay between the sender's transmission of 
a packet and when the ACK comes back.  This delay is the essence of the 
throughput problem.

> >    If possible, install a FOSSIL and use it in stead of Waffle's native ser
> > driver.  I believe that Waffle's DOS docs advise you to do this.
> 
> Yes, with the standard waffle uucico this is a good idea. I prefer BNU, but
> X00 works too from what I hear.

   I'm using X00, but am having problems using it with TeleMate's "FOSSIL" 
serial driver mode under DesqView; although X00 works like a charm with uucico 
under DV, TeleMate does a really lousy job of using it efficiently, resulting 
in problems if I download with TeleMate in a DV window.  Strangely, TeleMate's 
"Modem" driver (direct UART access) doesn't suffer from this problem, and 
uucico can receive a UUCP session via PEP and spoofing almost as fast as 
ZMODEM can download a .ZIP file!  I have to figure that it's TeleMate's FOSSIL 
interface... but I digress. 

> Not only does the author claim it, those of us using it will swear to it. :-)
> 
> I haven't used FX uucico with DV, but I have with Windows and OS2. There
> may be a bugaboo with DV alright, I've heard a number of people complain
> or comment on such things. But in a non-DV environment, you can't get a
> better uucico than FX. I'm a believer.

   I'd rather have slightly slower uucico in the background rather than faster 
uucico that monopolizes my machine.

> If anyone is interested in trying FX, it's available on halcyon, wuarchive,
> oak.oakland.edu, and other places by ftp. Look for FXUC03B.ZIP. It's also
> available via mailserver at fx-dist@dogear.spk.wa.us. Send mail to the
> server with the line   get fxuucico.zip  as the message body.

   I know that this isn't a particularly large file, but please don't 
encourage people to use mail based archive servers blindly.  Systems have gone 
down because of mbas use and others have been disconnected rudely by their 
feed because some uneducated user made use of them.  Sadly, any announcement 
of an available mbas service should be accompanied by an explanation of how 
these things work, why they're more damaging than direct FTP, and what the 
possible consequences of their use might be.

---  
  Geoffrey Welsh, 7 Strath Humber Court, Islington, Ontario, M9A 4C8 Canada
geoff@zswamp.uucp, [xenitec.on.ca|m2xenix.psg.com]!zswamp!geoff (416)258-8467
            Mathematics is the Mechano set with which we build our
       understanding of the Universe.  Unfortunately, I preferred Lego.