[net.sources.d] New Benchmark for UNIX Systems

duncan@ncrcae.UUCP (duncan) (04/08/86)

	[----------]

	Benchmarking of UNIX systems is still a young and growing endeavor
        for most of us in the computer industry.  There are several benchmark 
	programs available to both the system integrator and those trying to 
	compare system performance.  These programs consist of the BYTE,
	Whetstone, Dhrystone, AIM, Nelson Benchmarks, etc.  However, none of
	these programs provide a realistic test of the UNIX environment.  Async
	communications can not be tested by outputing some characters and
	ignoring the timing of input characters.  Disk subsystems can not be
	tested by reading and writing small sequential files and treating random
	files as not being relative to system performance.

	All of the subsystems of a computer must be analyzed together in order
	to get a realistic picture of system performance.

	The multi-user aspect of UNIX can best be tested by emulating what a
	typical user process(s) does.  It uses the CPU, TTY, and Disk resources
	all at the same time (relatively speaking).  With several user
	processes running at the same time, the system resources are allocated
	in fragmented packets of work.  It is these packets of work which a
	benchmark created here at NCR analyzes in order to evaluate system
	performance.

	The SCB (System Characterization Benchmark) was written in order to 
	test some of our new and existing systems.  It reduces a typical
	application to packets of CPU, Disk, and TTY work loads.  These units
	of work are then run concurrently to simulate a program's work load.
	The test runs a processor unit with varing numbers of Disk and TTY
	units.  Typically, the Disk load goes from 0 to 36 units and the TTY
	load varies from 0 to 120 units.  We here at NCR feel that this
	benchmark does a much better job of testing a system than most of the
	other currently available benchmarks.

	Now the good part!  This benchmark is FREE for the asking; all we ask
	is that you give NCR credit for creating it and that you not modify
	it without notifying us.  You can receive a diskette (IBM PC or NCR
	TOWER format), complete explanation of the benchmark (many pages), and
	operating instructions by sending me mail.  The complete benchmark 
	packet will be sent by regular surface mail.

	I would like to form a data base of results from this test, so if
	you request and use the SCB, I would appreciate getting a copy of
	your results.

	usenet mail address =   	decvax!mcnc!ncsu!ncrcae!duncan

	regular mail address =		Duncan McCorquodale
					NCR   E&M Columbia
					3325 Platt Springs Rd.
					West Columbia, SC.  29169