Tech Arsenal 1/Tech Arsenal (Arsenal Computer).ISO/tek-05/pclab303.zip/PCMAGTST

Text File | 1990-12-13 | 6KB | 125 lines

Overview

The PC Magazine LAN Benchmark tests exercise and evaluate
networks using tasks identical to those performed by real
applications.  Our goal in providing these tests is to give our
readers a way of reliably measuring the performance of their own
networks and to provide the LAN industry with a reliable set of
benchmarks.  However, please note that the results you get with
these and all other benchmark tests are only comparable to other
tests run under identical conditions.  The CPU power of the PCs
running this software is an often-overlooked factor that is
critical to the results you receive.

Test Structure

The concept behind these tests is that one network client PC runs
the Evaluation module of this program while all other PCs run a
selection from the Load module.  This architecture provides a
measurement of what one user would experience on a crowded
network and it is a valid way to evaluate overall network
performance.  The program on the PC running the evaluation test
will report its throughput in terms of kilobits per second.  If
you elect, the load programs will also report the amount of work
they did while the Evaluation program ran.  Together, these
statistics provide several ways to measure the work done on the
network and the throughput experienced by one station under
heavy load.

However, if you primarily want to measure server performance,
remember that just two or three fast PCs running Load programs
can use up a large percentage of the available bandwidth in any
10-16 megabit per second network.  In effect, the load stations
can clog the cable with data without materially impacting the
server.  It takes several 10-16 megabit per second LAN
connections to stress a modern high performance file server.
You will need multiple adapters in the file server and several
separate LAN segments to test the capability of a modern file
server with this or any other software.


File Sub-Menu

The File sub-menu lets you choose several options
including the size of the file you want the evaluation to use
and the name of the file where you want the results.  The
program will add consecutive tests to the results file and mark
them with a date and time, but you should use the spaces
provided to annote changes in the configuration of the LAN under
test.

The file size is important because many servers can cache very
large files.  The default, 1 MB, is cached for fast access in
many servers.  If you want to evaluate the effect of file size
on caching or hard disk performance, you can vary the figure.

This sub-menu contains the option to allow the Evaluation
program to control the start and stop of the load program.  You
can control these programs manually, but slaving them to the
Evaluation program produces a total picture of the work done by
the LAN during the evaluation period.

In networks with powerful servers, you should run the Evaluation
Program on a fast client station and provide a dedicated LAN
segment between the PC and the server.  Use multiple adapters in
the server to provide one or more additional links for the
client PCs running the Load programs.  A fast PCscan achieve
throughputs of over 4 megabits per second with a fast server, so
there isn't room for many client PCs on a single 10 megabit LAN
segment.

This program uses standard C programming language calls to read
and write data, sequentially and randomly, in blocks of 512
bytes, 4K bytes, and 16K bytes.  Note that most PC applications
use small block sizes and that the ability of the LAN hardware
and software to handle small blocks is an important factor in
this evaluation.  The results are presented in detailed form so
you can take whatever statistical results you find interesting.

Load Sub-Menu

The Load sub-menu provides you with a choice of several
different types of activities for the LAN client stations.  All
actions are the result of standard C programming language calls.
The Small Block test reads and writes one byte data blocks in a
temportary file.  This activity provides an exercise for the
network adapters and cabling.

The Hard Disk Load program reads data in a 10K temporary file,
but in servers with efficient caching schemes this activity will
be cached.

The Directory Access Load activity as close as a general purpose
benchmark program can come to simulating a program like a
database manager or other process running in the server.  It
runs the server's CPU through the endless task of examining
directory tables for a non-existant file name.  This load can
provide an interesting picture of the relationship between CPU
power and other factors in your file server.

The Printer Load activity simulates the load caused by multiple
print jobs transiting the print server.  The PC(s) running this
load activity must have their LPT1 port(s) re-directed to the
print server.  The print server must be active with a printer
attached.  The size of the print job will overwhelm the memory
in all typical laser printers and cause the print spool to feed
the printer slowly.  The nature of the print job assures that
you won't use much paper.

As the Load modules execute, they provide information on the
number of reads and writed they were able to complete through
the network.  If you record and compile this information, it will
provide you with a measurement of the total work done by the
networking system over a given period of time.