TESTNET.EXE Version 2.10 02-05-1991

By Scott Taylor
228 Gold Kettle Drive
Gaithersburg, MD. 20878
(301)-921-0437



**************************************************************************
*** THIS UTILITY IS GRANTED FOR USE BY THE GENERAL PUBLIC, BUSINESSES, ***
*** AND VENDORS. IT IS PROVIDED FREE OF CHARGE! ITS USE IS INTENDED TO ***
*** PROMOTE THE UNDERSTANDING AND OPERATIONS OF LOCAL AREA NETWORKS ***
*** AND HOW THEY RELATE TO DIFFERENT ENVIRONMENTAL CHARACTERISTICS. ***
**************************************************************************



TESTNET: This utility is intended to test the performance of Local Area
Networks within the realm of file I/O activity. It may also be used to test
stand alone PC's, or compare the two environments. This application has been
designed to allow for many different testing options, primarily to give more
real time results. This utility should also be used as a tool to gauge
performance while applications and tests are running on other nodes in a
network. TESTNET can also be use in conjunction with low-level packet analysis
programs, like The Sniffer or LANalyzer, for problem determination. It is
important to remember that benchmarking Local Area Networks is far from an
exact science. This test will however give you relative performance
statistics.

The program checks for the existence of Novell's NetWare. If Novell is
running, TESTNET will display the version of NetWare it is running under. Also
if the test is running under a NetWare environment the filename being used for
the test will have the same extension as the station number in the network.

In version 2.10 the all of the time sensitive routines, such as low-level file
I/O and timing, have been rewritten in assembler and optimized to provide
greater accuracy, and lower overhead, in calculating throughput. I tried to
eliminate as many non-essential instructions as possible in the timing
process. TESTNET is now in its 7th revision and 2nd generation. Most of the
recent changes have been made due to response from different LAN
Hardware/Software vendors.




RESULTS OF THE TESTS
(See EXAMPLE-3 For A Sample Test Result)

The valuable results of this test will show the total elapsed time in seconds
and total throughput in KB/sec transferred between the disk. An additional
feature is a display bar graph of total throughput, for each iteration, in
kilobytes per second. The bar graph has been enhanced since version 1.4. A
bright, or highlighted, bar represents the current throughput. A darker, or
shaded, bar extension represents the peak throughput for the currently running
test. Version 2.10 now displays a PASS counter beside the test filename. This
is used in conjunction with the REPEAT TEST mode. Every time the test repeats
the PASS counter will be incremented.



TESTING TECHNIQUES

NOTE: It is important to remember that test results will vary depending on
several of the following conditions. The speed of the CPU used for testing.
The speed of the disk or controller, and if caching is employed. The type of
network adapter and topology. The amount of CPU instructions used in the test.
Any TSR's, or memory resident utilities, that check interrupts or timer ticks.
And more importantly the type of test and parameters being used. There is no
precise method of determining how a system should perform. This test utility
will allow you to compare relative environments and operating systems. Please
consider all these factors when performing comparative tests!

Using a SEQUENTIAL file access method tests server disk and caching
performance. Using OVERLAID file access tests reflects more of the network
interface performance as well as local (workstation) cache techniques. Larger
record sizes are handled more efficiently by most operating systems
(especially DOS). Using larger record sizes (4K and up) demonstrates overall
throughput. Smaller record sizes demonstrates the operating systems ability
to deal with small record blocks. A good example would be to run this test
from a workstation local C: drive. Then run it again from that workstation but
to a network drive. Usually you will see a significant increase in
performance. Also try running the test with the VERIFY option on/off. Here is
another good example, especially for comparing different server hardware
(disk, memory, cpu). This is especially useful on Novell networks since the
server console displays performance characteristics. If using NetWare 2.10 or
greater, FCONSOLE will give some useful information. To compare hard disks
generate a file size, using Testnet, that is larger than the total memory
installed in the server. Then you can compare throughput times displayed by
Testnet. This way the file size is to large to fit in cache memory and tests
raw disk I/O.

To compare CPU's (as servers) or cache memory generate a file size, using
Testnet, that is approximately 512K smaller than the total memory installed
in the server. Then set Testnet to read the file, this will test the raw CPU
& Cache memory performance. Then you can compare the throughput displayed by
Testnet. It is best to use a large record size (eg. 32768), this will use the
least amount of CPU instruction overhead for performing the actual I/O. As the
test is in progress you can monitor the server console for peak "I/O PENDING"
and peak "% UTILIZATION".




MAIN MENU SCREEN FUNCTIONS
(See EXAMPLE-1 Below To View A Sample Of The Main Menu Screen)

MENU: The menu keys are the UP and DOWN arrow keys to highlight a menu option.
Press RETURN to select a highlighted option, then input the desired value for
that option. Instructions accompany each menu option when the RETURN key is
pressed. will exit TESTNET to the operating system.


RUNNING THE TEST: From the main menu starts the test in single station,
or stand alone mode.


RUNNING MULTIUSER TESTS: MULTI-STATION testing requires that SHARE be
installed, unless running Novell's NetWare. This is because aggregate results
are tallied in a file that uses record locking. Pressing sets the station
into the MULTI-STATION test mode. Simply press when the last station is
set into the MULTI-STATION test mode, then all stations activated will begin
testing. A new feature since version 1.4 is the ability to use identical test
parameters during MULTI-STATION testing. This is achieved by pressing
instead of . Pressing from the last node will cause all nodes to
start testing with that stations test parameters. Once the MULTI-STATION
testing begins each station will then show its average throughput to/from the
file server. All stations running in the MULTI-STATION mode must be able to
access the same root directory. A new feature in version 2.10 is the ability
to stop or restart all stations from a single node. Once MULTI-STATION testing
has begun, and repeat testing was selected, pressing the key on any
node will cause testing to stop on all nodes. TESTNET will then prompt with
"Continue Testing (Y/N)". Pressing will cause all other nodes to enter
MULTI-STATION testing mode again. You may then change test parameters if
desired and re-enter MULTI-USER test mode. Pressing the key will cause
all nodes to exit back to the operating system (DOS, OS/2, Windows, etc.)

Another new feature in version 2.10 is the display of aggregate and average
throughput. This function has a watch-dog timer built in that checks for nodes
that do not report. For example, if a node crashes or does not complete the
tests, the other nodes will consider that station dead and will continue
testing without it. The timeout period is two minutes after the last node
reported.

Version 2.10 now has a command line option, \M, to enter MULTI-STATION mode.
Using this will allow a test node to start up in MULTI-STATION mode without
any keystroke intervention. This is especially useful when starting network
stations from a batch file. Example: type 'TESTNET /M' .


DATA RECORDING: A new feature of TESTNET, since version 1.3, is a data
recording function. This is toggled on and off by pressing the key at the
main menu. The state of the recording function (Enabled or Disabled) is
displayed just above the menu instruction area. This will record historical
data to a file. This data file will be created in the root of the volume, or
drive, that you are testing on. You should take care to make sure that all
stations testing have access rights to the root of the volume or drive. The
data file is TESTFILE.DAT, and is used by all stations. The test data is
written to the file at the end of each test cycle. If Repeat Testing has been
selected, the data is written each time the test restarts. See EXAMPLE-5 at
the end of this document for a sample printout. This data can be used
comparatively. You can easily sort out the information by using the DOS SORT
utility. At the DOS prompt type just in the following 'TYPE TESTFILE.DAT|SORT
> LPT1:'. This will redirect the sorted information to the LPT1 printer port.
You can also replace the 'LPT1:' with a file name to have the sorted
information dumped into a file for viewing.


RESULTS REVIEW: The is used to view a record of all the tests that have
been performed at a particular station. This is a new feature for TESTNET
since version 1.4. Pressing will display all current test data (see
EXAMPLE-4 for a sample data screen). This is data that is kept in memory. Up
to 250 test results can be displayed. If more than 250 tests have been
performed, the remainder is ignored. This function is separate from the Data
Recording Function, because that saves data to a file. While viewing the test
data, the UP and DOWN arrow keys will scroll through the data. Pressing
will return to the Main Menu Screen. Pressing will print all of
the tests results in memory to a printer (Device LPT1 or PRN:).




MENU OPTIONS
(See EXAMPLE-1 Below To View A Sample Of The Main Menu)

REPEAT TESTING: This will allow testing to be repeated over and over again
until the key is pressed, then testing will discontinue. This option
will also work in MULTI-STATION test mode.


I/O DELAY CYCLE: This option is used to create run/stop periods. There are
actually two fields to this. The first is for defining how long transactions
will be exchanged, in seconds, before the test period is delayed. The second
option is for defining how long each delay cycle will last, in seconds. The
values are between 0 and 10 seconds. 0 indicated in any field will negate any
delaying at all. This option is useful for creating a pulsed environment that
more closely resembles a true network environment. You could set the run cycle
to 5 seconds and the delay period to 10 seconds for example. This means that
data is exchanged for 5 seconds, then stops for 10. And this is repeated until
all the iterations have been completed. This will not affect the time counter
in any way. This is an excellent way to measure the amount of degradation in
a hard working environment. Of course you could set many different operating
variables to produce different results depending on what type of network
conditions you wanted to resemble.


RECORD SIZE: TESTNET version 2.10 has increased the record size from 32768 to
65535 bytes. The record size is the size of each block to be read or written,
each time the disk is accessed for I/O. The values range from 1 byte to 65535
bytes. The most common physical block size for most disks is 4K (4096).
Although some operating systems are adjustable, such as NetWare 386.


REPETITIONS: Repetitions represents how many times the record, or buffer, will
be transacted (read or written). The values range from 1 to 10000 iterations.


I/O: This is the type of file I/O transactions that will occur. This
determines if you want to Read, Write, or Alternate (switch between Write &
Read).


FILE: This determines how the file will be accessed. Either Sequential access
(each record is appended to the end of the previous one), or Overlay access
(meaning each record overlays the previous one).


RECORD MANIPULATION: This is how the record size will be maintained. The
record will remain the same size (STD), it can increment from 1 to the value
specified (INC), it can decrement from the value specified down to 1 (DEC),
and it can be set to a random value between 1 and the value set (RAND).


Any questions, comments, or suggestions are welcomed !





*********************** EXAMPLE-1 (Main Menu Screen) ************************


TESTNET.EXE Version 2.10 02-05-1991 By Scott Taylor
Operating Environment| NetWare 386 V3.10 Rev. A


TESTNET - THE NETWORK & STANDALONE FILE I/O PERFORMANCE TESTER

DATA RECORDING DISABLED
REPEAT TESTING NO
I/O DELAY CYCLES <0 - 10> 0 \ 0 = Exit to system
RECORD SIZE <1 - 65535> 4096 = Start Testing
REPETITIONS <1 - 10000> 1000 = Start Multi-Station Test
I/O <1=READ 2=WRITE 3=ALT> 2 = Enable Data Recording
FILE <1=SEQUENTIAL 2=OVL> 1 = Display Test Results
RECORD MANIPULATION STYLE 4 <> & <> = Select Menu Option
<1=INC 2=DEC 3=RAND 4=STD> = Enter Data For Option













**************** EXAMPLE-2 (Multi-Station Mode Option Screen) *****************

TESTNET.EXE Version 2.10 02-05-1991 By Scott Taylor
Operating Environment| NetWare 386 V3.10 Rev. A


TESTNET - THE NETWORK & STANDALONE FILE I/O PERFORMANCE TESTER

DATA RECORDING DISABLED
REPEAT TESTING NO
I/O DELAY m
RECORD SIZ
REPETITION MULTI-STATION TESTING MODE ation Test
I/O <1=REA cording
FILE <1=SE esults
RECORD MAN This mode is to allow multiple stations nu Option
<1=INC 2=D to run the I/O tests at the same time. For Option

All stations to participate in the test
should be set into this mode (pressing
from the main screen). To begin
testing, press from one of the work
stations activated. All nodes will then
start testing. Pressing will cause
all nodes to begin testing with the same
set of test parameters. WILL ABORT!


****************** EXAMPLE-3 (Sample Testing/Results Screen) ******************

TESTNET.EXE Version 2.10 02-05-1991 By Scott Taylor
Operating Environment| NetWare 386 V3.10 Rev. A


TESTNET - THE NETWORK & STANDALONE FILE I/O PERFORMANCE TESTER

DATA RECORDING DISABLED
REPEAT TESTING YES
I/O DELAY CYCLES <0 - 10> 0 \ 0 -FILENAME- -PASS-
RECORD SIZE <1 - 65535> 65535 TESTFILE.001 #001
REPETITIONS <1 - 10000> 100
I/O <1=READ 2=WRITE 3=ALT> 2 NODES IN TEST # 1
FILE <1=SEQUENTIAL 2=OVL> 2 NODES REPORTED # 1
RECORD MANIPULATION STYLE 4 AVERAGE KB/SEC # 219.77
<1=INC 2=DEC 3=RAND 4=STD> AGGREGATE KB/SEC # 219.77
[ COMPLETED ]

TIME = 29 I/O COUNT = 100 TOTAL BYTES = 6553500 KB/SEC = 219.77

KB/
SEC
0 20 40 60 80 100 120 140 160 180 200 220 240 260

Total Seconds = 29.12


************* EXAMPLE-4 (Sample Display Of Cumulative Test Data) *************

TESTNET.EXE Version 2.10 02-05-1991 By Scott Taylor
Operating Environment| NetWare 386 V3.10 Rev. A

NODE PASS OPERATION I/O TIME BUFFER I/O COUNT TOTAL BYTES KB/SEC

001 001 WRITE SEQ 9.12 4096 500 2048000 219.28
001 002 WRITE SEQ 9.12 4096 500 2048000 219.28
001 003 WRITE SEQ 9.12 4096 500 2048000 219.28
001 004 WRITE SEQ 9.12 4096 500 2048000 219.28
001 001 READ SEQ 8.96 4096 500 2048000 223.31
001 002 READ SEQ 8.79 4096 500 2048000 227.50
001 003 READ SEQ 8.74 4096 500 2048000 228.93
001 001 ALTERNATE OVL 8.90 4096 250 2048000 224.69
001 002 ALTERNATE OVL 8.96 4096 250 2048000 223.31
001 003 ALTERNATE OVL 8.96 4096 250 2048000 223.31








[ ESC-Key To Exit ] [ F1-Key To Print ]




********** EXAMPLE-5 (Sample Printout Of Recorded Data File Results) **********

TYPE TESTFILE.DAT

NODE PASS OPERATION I/O TIME BUFFER I/O COUNT TOTAL BYTES KB/SEC
001 001 WRITE SEQ 6.04 32768 100 3276800 529.45
001 001 READ SEQ 5.99 32768 100 3276800 534.31
001 002 READ SEQ 5.99 32768 100 3276800 534.31
001 003 READ SEQ 5.99 32768 100 3276800 534.31
001 001 ALTERNATE OVL 21.76 65535 100 13107000 588.27
001 002 ALTERNATE OVL 21.70 65535 100 13107000 589.76
001 003 ALTERNATE OVL 21.70 65535 100 13107000 589.76