Disktest raw man text
Copyright (c) International Business Machines Corp., 2001


This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.

This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.

You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.

Please send e-mail to yardleyb@us.ibm.com if you have
questions or comments.

Project Website: TBD
Process this file with
groff -man -Tascii disktest.1

$Id: disktest.1,v 1.5 2008/02/14 08:22:25 subrata_modak Exp $

disktest - Test tool for exersizing disk devices disktest [-q] [-Q] [-r] [-w] [-E cmp_length ] [-a seed ] [ -A action ] [-z|c|n|f fixed_pattern ] [-B sBLK[:eBLK] | -C cycles ] [-d ] [-D r%:w% ] [-F] [-h heartbeat ] [-K threads | -L seeks ] [-P TXPRCA ] [ -m ] [-N sectors ] [-R retry[:retryDelay] ] [-o offset ] [-s sLBA[:eLBA] ] [-S sBLK[:eBLK] ] [-t delayMin[[:[delayMax]][:ioTimeout]] ] [-T seconds ] [-p r | R | l [ u | d ] ] | L [ u | d ]] [-I [ d ] r | b | f [ s [ delay ] ] ] filespec Disktest does repeated accesses to a filespec and optionally writes to, reads from, and verifies the data. By default, disktest makes assumptions about the running environment which allows for a quick start of IO generation. However, Disktest has a large number of command line options which can be used to adapt the test for a variety of uses including data integrity, medium integrity, performance, and simple application simulation. Disktest will use the device specified by filespec. If no option is specified otherwise, disktest will attempt to determine filespec type. Fully qualified path must be give when filespec is not a normal file. This will help to determine it's type. Most options have multipliers that can be used to specify larger amounts. The following are a list of these multipliers. These can be used on options such as -B, -L, -N, -S, -s. The time options also have multipliers. These can be used on options such as -h and -T. Displays a short description of the command line options and exits normally. Use seed for all random number generation when constructing blocks of pseudo-random data and random seeks. By default, seed is set to the process id number. To reproduce a previous test run, use the process id number outputted to stdout. These options are used to modify the default behavor during IO runtime. The following options can be used for these modifications. Set the size of the data block transfer. If only sBlockSize is specified, then the transfer length is always a fixed length of sBlockSize. If eBlockSize is specified, then the transfer length will be randomly chosen between sBlockSize and eBlockSize. The transfer size will always be a multipile of the sector size. If either parameter is greater then 256, then the value will be integer divided by the sector size default, which is 512 bytes. If either parameter is preceded by a 'k', i.e. 8k, then the value will be multiplied by 1024. Otherwise, the parameters will be taken literally. The default block size is (1*sector size) or 512. Note that O_DIRECT , no filesystem buffering, and some file system may not be able to perform accesses as small as 512 bytes. This will result in an IO failure with a transfer length of -1. Use a counting sequence for the bytes within to each block. The count starts at 0 and increments to 255 then begins again at 0. Each sector is filled with two of these sequences. Run until the total number of cycles are complete. When cycles is set to zero, disktest will run until killed. The -L or -T option is used to specify how long each cycle will run for. If neither -T or -L is specified, the cycle length with be calculated to the number of seeks based on the device size, -N, divided by the transfer size, -B. If -C is not specified, only a single cycle is run. Force disktest to dump to stdout the amount of data at the location specified by the other command-line option, i.e. -d -s50 -B 32k will dump 32768 bytes of data to stdout starting at LBA 50. The data is formatted into lines of 16 bytes with the location offset and ASCII equivalent. Duty cycle used while reading and/or writing. For example, -D 20:80 would cause disktest to generate a read 20% of the total run time and generate a write 80%. If only read or write is give then the percentage is always set to 100 for the specified option. If the total percentage does not add up to 100, i.e. -D 20:70, then disktest will split the remaining percentage, resulting in 25% reads and 75% writes. Turn on error checking. Data read from filespec will be checked for correctness up to the number of bytes specified by compare_length. If compare_length is 0 then the block size is used as the compare length. By default, data read is not checked for errors. Use a data pattern consisting of a fixed value. fixed_pattern can be entered using decimal, any number not starting with a 0, octel, any number staring with a 0, or hexadecimal, any number and [A-F] starting with 0x. The value can be no greater +/- 2^63, in size. used to specify that the filespec is actually a file listing the targets that disktest should operate on. This allows disktest to run to multiple targets using the same options from a single command-line. Performance data will be sent to stdout every heartbeat seconds. During a linear test, -pL, only heartbeat statistics for the current operational cycle will be displayed. The default is to only display performance data at the end of the test, which is cumulative for all IO performed throughout the test. Set the data transfer type to IO_type. Valid IO types are r (raw), b (block), or f (file) I/O. These options are case sensitive. The r (raw) type is used when binding a block device to a raw device, see raw(8). Disktest will align it's buffers correctly to support raw devices. The b (block) type is used when block IO is desired. The buffer_cache will be used during testing. Buffer alignment is not required for this type of IO operation. The f (file) type is used when accessing a file. If the file does not exist then it will be created. If the file exists, then it will opened; see O_CREAT in open(2) for more details. Access to the file is performed through the file system that the file is stored on. Adding an S modifier to the F (file) type opertaions will force an fsync(2) to occur on every write. Adding d will open with the O_DIRECT flag set. If this option is used, then I/O is limited to being aligned to the file systems block size. When transferring to a block device w/o a file system, then alignment is to 1k. These limits have been verified with the 2.4.9 kernel and the o_direct patch from AA. Adding s sync_interval Specifies that a sync should occur at sync_interval number of write IO operations. The default is to sync on every IO. Disktest will report a failure if filespec does not match the IO_type specified. If no type is specified, then disktest will attempt to determine the file type by using stat(2). Set the number of test threads to threads. Each child can read or write based on the specified criteria. The default number of test threads is 4. Total number of seeks to occur during testing. This option specifies the exact number of times a seek occurs on a resource. By default disktest will calculate the number of seeks by taking the difference between the start block and the stop block. If the difference is 0 then the default is 1000 seeks. This option will add the lba, pass count, seed, cycle start time, hostname, and target to each LBA as header data before any IO operation occurs. The mark replaces the first n bytes of data in each LBA. This option will override the cycle start time in the -m option with the specified value. This is useful when you are writing in one distest instance and read/verifing in another so that the mark data can be unique and also deterministic. Use a data pattern that consists of the the lba number. An lba value of up to a 64b can be stored. The 64b value is repeated to fill the transfer buffer. Set the number of available sectors to num_secs. If no num_secs is specified, and the size of the device can be determined, then the number of sectors, as reported by the device is used, otherwise, the default number of sectors is 2000. This option specifies the LBA offset to shift all alignment of IO by. For example, if a test is to perform a full stride write on a storage device, and the os and/or storage device offset the strides by a number of LBAs, this parameter can be used to set that offset, so that IO is aligned to the stride on the storage device. By default the offset is set to zero. Set the pattern of seeks to seek_pattern. Valid patterns are l (linear interleaved writes/reads), L (linear writes then reads), R (random), r (random interleaved writes/reads). Linear may also specify what happens when the last block is reached. Option u specifies that the test should start back at first block after reaching the last block. d specifies that the test, after reaching the last block, should start at the last block and go to the first block. The default extra option for linear is 'u'. The default seek is random. Record performance statistic to stdout. Perf_opts is a string of characters representing which statistics should be reported. The possible options are: Suppress all the 'INFO' level messages that are send to stdout. This includes all the assumption messages the disktest will print as it finds that the option was not specified in the command line arguments. Suppress header data from messages that are send to stdout. Read from filespec. This is the default option if -w or -r are not specified. -E must be specified if data integrity checking is desired. Specifies that on a seek or transfer error, the IO should be retried. retry Specifies the number of retry attempts that should be made. retryDelay Specifies the amount of time delay (msec) before the retry occurs. If no retryDelay is specified, then retries will occur immediately. By default, IO errors are not retried. Set the starting test block to start_block and the ending test block to end_block. By default, start_block is 0 and end_block is 2000. If end_block is not given, and the size of filespec can be determined, then end_block is set to the volume capacity reported by the device divided by the transfer length. This option can only be used when there is a fixed transfer length. The range given is inclusive, so if -S0:10 is specificed, this will be 11 blocks. The -S and -s options are exclusive. Set the starting test LBA to start_LBA and the ending test LBA to end_LBA. By default, start_LBA is 0 and end_LBA is 2000. If end_LBA is not given, and the size of filespec can be determined, then end_LBA is set to the volume capacity reported by the device. This option can only be used when there is a fixed transfer length. The range given is inclusive, so if -s0:10 is specificed, this will be 11 LBAs. The -S and -s options are exclusive. Wait delayMin milliseconds between each IO. This is used when attempting to simulate a static load from an application that has some known proccessing time between IO operations. delayMax can be added to specify that a per thread random IO delay should be used, between delayMin and delayMax. When combined with multiple threads, -K, these can be used to model IO load for simulating application processing after IO. By default, disktest will issue as many IO requests as possible, delayMin/delayMax is set to zero, which may over drive some disk subsystems when multiple hosts running disktest are attached to the same disk subsystem. If no IOs complete in ioTimeout seconds, then disktest will consider the test to fail. The default is now 60 secs, which means that if there are no IO operations to a target from any thread that complete in 60 secs then the test will stop with a failed status, and an ERROR message stating the there is a possible hung IO condition, if it is a true hung IO condition, then disktest IO threads will not terminate with a non-preemtable kernel, and the only error message with be from the ioTimeout ERROR message. To disable this feature, set the io timeout to 0, which means that the IO timeout time will never be reached which is how disktest operated before this feature was added. The minute, m, hour, h, and day, d, multipliers can also be used on these perameter. The following are examples of -t usage. Run until runtime seconds have elapsed. Runtime must always be greater than zero. -T and -L are exclusive of one another. The version information will be displayed and disktest will exit normally. Write to filespec. Data will be written as fast as possible and not read back to check for data corruption. can be combined with -r option to do read/write testing and -E to perform data integrity checking. Use a randomly generated data pattern based on the seed for the bytes within to each block. The data pattern is random for the first 512 bytes, one LBA. The pattern is then repeated for each LBA after creating a pseudo random data pattern across the given filespec. This is done for two reasons. One, it saves on the memory foot print size need and time required to generate the data, and two, an LBA is the smallest unit of work disktest operates on. Therefore, disktest can maintain the ability to do data checking, random block size transfers, and random block offsets when using random data. ./disktest None. The following are some examples on how to use the options in disktest to create different types of workloads. Please use these as a guideline to get started.