farber%pcpond.pc.udel.edu@Louie.UDEL.EDU (David Farber) (06/11/86)
To: farber@pcpond.PC.UDEL.EDU
Subject: Net remarks
Date: Sat Jun 7 22:48:21 1986
From: doug%sco.uucp@cfg.PC.UDEL.EDU
>From doug Sat Jun 7 22:44:43 1986
To: ihnp4!seismo!men1!andys
Subject: Your comments
Cc: doug
Date: Sat Jun 7 22:44:37 1986
Andy, we are very pleased you have had such a positive experience
with your XENIX system. Your comments on its solidity, performance,
and ease of use are the sort of thing we look forward to hearing.
I would like to clarify a couple of things alluded to in your summary.
Serial I/O performance: You are correct that dumb cards do present a
serious and difficult load for any computer. An interrupt per character
at 9600 baud is about 1000 interrupts per second. A normal UNIX style
driver makes a context switch for each interrupt and this results in almost
all CPU cycles being used up. SCO has invested enormous effort to build
a driver that minimizes context switches and thus improves performance
dramatically. Unfortunately we do not currently support the Bell serial
board with this driver. They have provided their own and it does not
have the same level of performance. I think you would find if you used
one of the boards supported by SCO you would see a marked improvement.
C compiler problems: We have been distributing a beta release of a newer
version of the C compiler, it will be incorporated in the 2.2 release of
XENIX later this year. We have held off until we could extensively test it.
Any SCO customer may get a copy by simply calling our toll free support
number and requesting it. There is a known limitation in the compiler
because certain storage classes must go into DGROUP, which is limited
to 64kb. The main problem here is initialized global data. This is
an unfortunate restriction and we are working with Microsoft to eliminate
it. While I sympathize with the annoyance of working around this situation
it should not be ignored that the Microsoft C compiler is an extremely
powerful and flexible compiler in general. It produces better code then
any other 286 compiler we know of and has the added flexibility of DOS
compatibility as well as a complete DOS cross development environment.
In addition it has a number of extensions that make writing high performance
programs on the 286 considerably easier then with other systems.
As to some of your other suggestions they are generally positive ideas to
improve the product. We will be adding a complete permuted index to make
using the documentation easier, we are considering on-line man pages now
that disks are typically bigger then 20 MB, we will provide support for
disks not covered by ROM.
My largest concern is your quickness to praise competing systems that you
have not evalutated as carefully may mislead others. SCO is more then
willing to be objectively compared side by side with any other version
of UNIX that runs on the AT as long as you fairly compare: performance,
stability, reliability, range of peripherials supported, quality of
support, overall quality and completeness of documentation, I/O performance
with supported peripherials (especially serial ports, try COM1 and COM2 for
comparison), commercial applications available, etc. We know our product is
not perfect and we are continually improving it, largely based on informed
suggestions such as yours. We do not believe that any other product on
the market will outperform SCO XENIX in an objective evaluation.
Thanks for using XENIX, please keep your suggestions comming.
Doug Michels
Vice President