Path: utzoo!attcan!utgpu!jarvis.csri.toronto.edu!cs.utexas.edu!sun-barr!newstop!texsun!texbell!vector!attctc!mjbtn!raider!wlk
From: wlk@raider.MFEE.TN.US (Wolf Kozel)
Newsgroups: comp.unix.xenix
Subject: Smokin' Serial Port Boards
Keywords: Arnet Specialix Equinox Benchmarking Serial Port Boards
Message-ID: <157@raider.MFEE.TN.US>
Date: 6 Dec 89 21:12:07 GMT
Organization: RaiderNet Public Access, Murfreesboro, TN
Lines: 37

Greetings!

Serial Port cards speed is a big factor in choosing one along
with cpu usage.  If you're using Unix/Xenix, there are some
factors that you should consider.

1) Speed improvements will come from software, not hardware.
   According to an Infoworld article, there seemed to be a
   bottleneck itself in Xenix with serial port cards.  Testing
   a board in DOS doesn't mean the same performance in SCO.
   Computer Design had an article on some serial io chips that
   can transfer data at tremendous speeds.  The catch is the
   cost.  Three hundred dollars instead of $20 we're paying
   now.  Would anybody pay this?  One engineer said a rule
   of thumb was tripling material cost -- which means an
   eight port board would cost around $2200.

2) Downloading the i/o portion of the kernel to the card, while
   seductive, results in only a 5-10% increase in speed.  This
   is not a trivial task.  There are probably bigger wins out
   there.  

3) By tweaking driver code you can optimize it.  We've increased
   the speed of our driver by almost 100% in benchmarks.  But
   when we ran the driver on a 'Real-World' system with users
   and windows and stuff, there wasn't a noticeable difference.
   There was a big difference when there was just one person on
   it, however.  

   Co-workers at Arnet tell me that not too many people seem
to really care about benchmarks.  They cite Fall Comdex where
only a couple came up to our booth to ask about them.  It seems
to popular on the net, I wonder it that is indicative?  

   Thanks for reading this,
      Wolf Kozel  (Arnet System Administrator)