home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
Network Supervisor's Toolkit
/
Network Supervisor's Toolkit.iso
/
perform
/
netframe
/
readme.dx
< prev
next >
Wrap
Text File
|
1996-07-10
|
10KB
|
200 lines
NetFRAME Systems Inc.
DX
A Disk Exerciser for NetWare 386 (TM)
WHAT IS DX?
DX (Disk eXerciser) is a portable performance analysis tool. It can be run
on systems that run Novell's NetWare 386 (TM). DX was developed at
NetFRAME Systems Inc., in response to a need to analyze and characterize
physical disk IO performance, without file system caching effects,
over a variety of operating conditions.
DX is a NetWare Loadable Module, or NLM. It can be loaded and run from
the operator's console of a NetWare 386 system. It can provide a variety
of system configuration options in examining physical disk IO throughput.
It is portable and easy to use, providing users a variety of reproducible
conditions.
BEFORE YOU START
DX is provided on either a 3.5" or 5.25" floppy, along with this file. Its
filename is DX.NLM. You must have a system capable of running NetWare 386,
a copy of NetWare 386, Rev 3.0 or higher, and at least one hard disk.
DX should not be run on any disk that has volume SYS mounted.
This is because DX can do destructive writes on any location in a NetWare
partition. In general, if you wish to run DX on a system with useful data
on the disks, you must backup all of the data before running DX, and later
restore the data.
DX can only be run on disks or drives that have a NetWare partition on
them. If you wish to run DX on a disk that doesn't have a NetWare partition,
you must create one using the INSTALL utility provided with NetWare. See
the appropriate Novell documentation to use this utility.
HOW TO RUN DX
First, bring up NetWare 386 on your system. Do not mount SYS or any other
volume(s). Be sure to load the appropriate disk driver(s) to exercise your
disks. Other NLMs could be loaded (e.g. INSTALL, MONITOR) as required, but
they are not necessary to run DX. Be sure all useful data on your drives
are backed up. Either download the dx.nlm into your DOS partition, copy
into the system directory of the Netware 386 volume, or load from a floppy
drive. At the operator's console, type LOAD DX. DX will clear the screen
and ask a series of questions:
1. Exerciser Access Pattern [S=Sequential R=Random]? S
The answer to this question determines how DX will access the disk-
sequentially or at random. In sequential access, DX will start requesting
data at the beginning of the drive, and increment the block address for
each successive operation. When it reaches the end of the drive, it will
wrap around and start at 0 again. In random access, a random number
generator with a uniform distribution is used to generate a block address
ranging from the beginning to the end of the drive.
Sequential access tests measure a system's capacity to read or write
large files. Random access tests, with uniform distribution, are useful in
examining a system's ability to handle heavy IO demands with poor disk locality.
2. Exerciser test type [R=Read W=Write]? R
The answer to this question determines if DX will perform all reads or all
writes for the duration of the test. Note that data
read from drives by DX is discarded. Data written to drives is not
subsequently verified by this program. Use of the NetWare parameter "Enable
Disk Read After Write Verify" can be used to provide write verify performance
measurement, in concert with DX's write option. However, not all disk
drivers implement this feature, even though the option is turned on. Use care
in examining the results before assuming the feature is supported. There
should be a noticeable drop in write throughput when write verify is on and
supported. The default operation is Read.
3. Number of kilo bytes <KB> per IO [1-60]? 8
This answer instructs DX on how big each physical IO should be--1 to 60KB
long. The default is 8KB per IO. Small IO operations (1KB) can give an
indication of the maximum IOs per seconds a controller/drive(s) combination
can sustain. Medium size IOs can give an indication of how a common NetWare
system may perform (many NetWare systems are configured with 4-16KB
cache buffer sizes, which corresponds to the blocksize of the
disk request). Large IOs can give an indication of the data throughput that
is sustainable, since large IOs usually minimize controller to drive protocol
overheads.
4. Number of concurrent IO to queue per drive [1-64]? 3
DX is able to build and maintain a queue of requests for each drive before
calling the disk driver to act on those requests. This number lets DX know
how deep to keep this queue. This feature is useful for intelligent
drivers and/or controllers that are capable of issuing multiple IOs,
for elevator algorithms that take advantage of sorts to improve disk access
time, etc. The default queue depth is 3.
5. Screen Update Interval in seconds [1-60]? 5
This tells DX how often to calculate and display statistics on the console.
The default is 5 seconds.
6. Exercise Generic Drive 1-0-0? y
For each drive in the system with a NetWare partition, DX will ask if you want
to include that drive in the test. This is useful for excluding certain
drives with data you want untouched, or to characterize your system on a disk
by disk, controller by controller, or system wide basis. The default answer
to these questions is yes, so care is required. The format of the drive
number is Controller number, Device ID, LUN. Also note that the search for
disk drives will end when DX encounters a drive without a NetWare partition.
THE DX DISPLAY
After questioning you about all the drives with NetWare partitions, DX will
clear the screen and start exercising the disk(s) you specified in the manner
you requested. After the screen update interval, it will display the statistics
it has gathered so far. Here is a sample display:
--------------------------------------------------------------------------------
NetFRAME Disk Test: Sequential Read 32KB/IO with 3 IOs queued Page 1 of 1
Disk KB/sec. IOs/sec Ave. KB/s Total IOs Errors
Generic Drive 1-0-0 812 25 752 9976 0
Aggregate Disk Performance 812 25 752 9976 0
Server Utilization 7%
Q to quit or page number to view
--------------------------------------------------------------------------------
NOTE: The numbers supplied above are provided for reference only and do not
reflect the physical disk performance of any system.
The top line identifies DX, the type of test it is running, and the page
being displayed. DX can display statistics on very large systems. It has been
tested on a system with 32 drives. Each page of a DX display has statistics
for up to 16 drives. Aggregate statistics are shown on every page.
To display the information on any page, type in the page number on the
keyboard at the operator's console. After the next display interval, that page
will be shown until a new page number is typed, or you exit DX.
To exit DX, type "q" (the letter only). DX will ask you if you really want
to exit. Type y or n at the prompt, which will appear at the bottom line of
the display. The default is yes.
You may also do other things while DX is running. You may use the
<ALT-EXC> or <CTL-ESC> key combinations to cycle through any other NLMs you
have running, or to the console. However, other activities on the system
could perturb your results.
The second line consists of column heading, to identify which statistics
are collected in that column. The Disk column identifies which drive is
being tested. The KB/sec column identifies how many kilobytes per second were
transferred during the last display interval. The IOs/sec column shows how
many physical IOs per second were completed during the last display interval.
The AVE. KB/s column displays the average data transfer rate since the beginning
of the test. The Total IOs column shows the number of completed physical IOs
since the beginning of the test. The Errors column details the reported
physical IO errors since the beginning of the test.
DX then displays statistics for up to 16 drives per page. It also
sums the statistics for all tested drives on the AGGREGATE Disk Performance
line. These statistics are shown on all pages. Finally, the CPU utilization
for the server is displayed. CPU utilization is the percentage of processor
time used marginally to process the benchmark results and mostly to execute
the disk IO requests. The remainder (unused CPU cycles) is the processor time
available to perform other functions on the system.
CAVEATS
To change test parameters, you must quit the program and reload it.
There is no explicit way to capture DX displays to a file or files.
DX only displays information about completed IOs.
DX has no provision for gathering statistics on a per controller basis.
DX has no provision for doing 512 byte (a common disk sector size) IOs.
DX has no provision for operating only on part of a NetWare partition.
DX tests can be destructive to data.
DX is not a capacity planning tool and does not predict total system demands
for particular customer environments.
When running both DX and NX, be sure to load both NLMs before starting
either. If you don't, server CPU utilization percentages will be
in error. This is due to the way NetWare 386 handles the loading of NLMs
in relation to server CPU utilization.
DX is provided as is, with no warranty, explicit or implied. NetFRAME Systems
Inc. assumes no responsibility for any information provided by DX, nor any
effects, direct or indirect, on your system. DX has been tested on NetFRAME
systems, and on a limited number of ISA, EISA, and MCA systems, and has shown
no deleterious side effects to systems under test. However, DX may exhibit
side effects on your system.
TRADEMARKS
Novell, NetWare, NetWare 386, NetFRAME, and NetFRAME Systems, are the trademarks
of Novell, Inc. and NetFRAME Systems Inc.
(c) 1990 NetFRAME Systems Inc., 1545 Barbar Lane, Milpitas, CA 95035