Path: utzoo!utgpu!water!watmath!clyde!rutgers!ames!think!husc6!bbn!uwmcsd1!ig!agate!ucbvax!hplabs!oracle!rbradbur
From: rbradbur@oracle.UUCP (Robert Bradbury)
Newsgroups: comp.sys.ibm.pc
Subject: Re: XENIX VS Micoport
Summary: Microport speed - lets find the source of the problems
Keywords: Microport, XENIX, speed, profiler, sar
Message-ID: <242@oracle.UUCP>
Date: 14 Feb 88 01:38:37 GMT
References: <16217@watmath.waterloo.edu> <166@bhjat.UUCP> <366@igloo.UUCP> <434@cimcor.UUCP> <391@igloo.UUCP>
Reply-To: rbradbur@oracle.UUCP (Robert Bradbury)
Distribution: comp.sys.ibm.pc
Organization: Oracle Corporation, Belmont, CA
Lines: 72

Over the last month I've seen alot of bad press about the speed of
Microport UNIX.  It started with the Unique article which in my
opinion gave an excessively negative imaage of Microport UNIX and
has continued here.  The basic issue seems to revolve around the speed
of the two systems as related to disk I/O and serial port I/O.
Unfortunately people seem to be describing symptoms and complaining
instead of actually doing some hard research into the nature of
the problems.

RE: hard disk speed.  Has anyone done the research to determine the
proper interleave factor for the hard disks?  Has anyone determined the
interaction between the software interleave (specified during when
making UNIX file systems) and hardware interleave handled by the
controller?  Exactly what are the I/O transfer rates you can
expect under UNIX?

RE: serial I/O ports.  I'm under the impression that standard PC
I/O ports do character at a time I/O.  That means you have to
take an interrupt for every character into or out of the machine.
What is the interrupt overhead on the UNIXes?  I once managed a
VAX with a DZ multiplexor (which did character at a time I/O) and
we could routinely bring the machine to its knees by running a
couple of lines doing medium speed I/O.  Unless you have very fast
machines with low interrupt overhead you cannot expect to do
much I/O with character at a time devices.  If you want to do
alot of I/O and not bog down the CPU go get a board like the
DIGIBOARD COM/8I with another processor (80186) and 8K I/O buffers
to handle the I/O.

I'm not saying that XENIX is not faster than Microport, after all SCO/
Microsoft has had 2-3 more years to refine the serial I/O port driver
and hard disk interleave factors.  The fact that most XENIX development
was done of VERY SLOW machines gave the developers an incentive to
improve those features.  Now people have so much CPU power available
to them that they rarely put themselves in an environment which degrades
peformance sufficiently to provide an incentive to improve things.

One of the advantages of Microport over Xenix is that it provides tools
to analyze the performance of the machine (SAR and the system profiler).
SAR works.  A sequence like the following:
	/usr/lib/sa/sa1
	 - execute command(s)
	/usr/lib/sa/sa1
	sar -y			- give tty I/O statistics
	sar -b			- give disk I/O statistics

These would give us (and Microport) an idea of how heavily the system
is being used and what the bottlenecks might be.

The system profiler would provide even more information about exactly
which functions in the UNIX kernel were consuming the most time.
(Unfortunately it appears the profiler is not operative in my current
 2.2 release of 386 Unix -- -2 for Microport).

As these features are not available under Xenix (remember the Xenix
kernel is based on System III) you have no assistance in locating
the source of system bottlenecks.

I would be interested in some HARD facts about how many terminals
you can run at what speeds with and without console I/O, hard disk
interleave factors, interrupt overhead, I/O bandwidth, etc. for
machines running Microport UNIX and XENIX.  Then we could give
the O.S. and the machines a "FAIR" comparison.

Should we develop a set of programs which can measure these things?
I'm willing to provide the database maintenance for the results.


-- 
Robert Bradbury
Oracle Corporation
(206) 784-9726                            hplabs!oracle!rbradbur