Accounting Utilities Manual


Next: , Previous: (dir), Up: (dir)

Welcome to the GNU Accounting Utilities! If you don't have a clue about the accounting utilities, read the introduction. For specfic information about any of the applications, see the chapter with the program's name.

This is Edition 6.6.2 of the documentation, updated 19 August 2015 for version 6.6.2 of the GNU Accounting Utilities.


Next: , Previous: Top, Up: Top

Preface

Way back a long time ago, Thompson and Ritchie were sitting opposite one another at the commissary, sipping coffees and discussing their evolving behemoth.

“This behemoth of ours,” said Ken, “is becoming rather popular, wouldn't you say?” “Yes,” said Dennis. “Every time I want to do a compilation, I have to wait for hours and hours. It's infuriating.” They both agreed that the load on their system was too great. Both sighed, picked up their mugs, and went back to the workbench. Little did they know that an upper-management type was sitting just within earshot of their conversation.

“We are AT&T Bell Laboratories, aren't we?” the upper-management type thought to himself. “Well, what is our organization best known for?” The brill-cream in his hair glistened. “Screwing people out of lots of money, of course! If there were some way that we could keep tabs on users and charge them through the nose for their CPU time...”

The accounting utilities were born.

Seriously though, the accouting utilities can provide a system administrator with useful information about system usage—connections, programs executed, and utilization of system resources.

Information about users—their connect time, location, programs executed, and the like—is automatically recored in files by init and login. Four of them are of interest to us: wtmp, which has records for each login and logout; acct, which records each command that was run; usracct and savacct, which contain summaries of the information in acct by user and command, respectively. Each of the accounting utilities reports or summarizes information stored in these files.

ac
prints statistics about users' connect time. ac can tell you how long a particular user or group of users were connected to your system, printing totals by day or for all of the entries in the wtmp file.
accton
turns accounting on or off.
lastcomm
lists the commands executed on the system, most recent first, showing the run state of each command. With last, you can search the acct file for a particular user, terminal, or command.
sa
summarizes the information in the acct file into the savacct and usracct file. It also generates reports about commands, giving the number of invocations, cpu time used, average core usage, etc.
dump-acct
dump-utmp
display acct and utmp files in a human-readable format.

For more detailed information on any of these programs, check the chapter with the program title.

A Note on File Names and Locations

The wtmp and acct files seem to live in different places and have different names for every variant of u*x that exists. The name wtmp seems to be standard for the login accounting file, but the process accounting file might be acct or pacct on your system. To find the actual locations and names of these files on your system, specify the --help flag to any of the programs in this package and the information will dumped to standard output.

Regardless of the names and locations of files on your system, this manual will refer to the login accounting file as wtmp and the process accounting files as acct, savacct, and usracct.

Support for Multiple Accounting File Formats under Linux

The detailed format of the acct file written by the Linux kernel varies depending on the kernel's version and configuration: Linux kernels 2.6.7 and earlier write a v0 format acct file which unfortunately cannot store user and group ids (uid/gid) larger than 65535. Kernels 2.6.8 and later write the acct file in v1, v2 or v3 formats. (v3 if BSD_PROCESS_ACCT_V3 is selected in the kernel configuration, otherwise v1 if on the m68k architecture or v2 everywhere else).

Since version 6.4 the GNU accounting utilities on Linux systems are able to read all of the v0, v2 and v3 file formats (v1 is not supported). Thus you do not need to worry about the details given above. You can even read acct files where different records were written by differently configured kernels (you can find out about the format of each entry by using the dump-acct utility). In case you ever need to convert an acct file to a different format, the --raw option of dump-acct does that together with the new --format and --byteswap options that determine format and byte order of the output file.

Multiformat support under Linux is intended to be a temporary solution to aid in switching to the v3 acct file format. So do not expect GNU acct 6.7 to still contain Multiformat support. In a few years time, when everybody uses the v3 format, the ability to read multiple formats at runtime will probably be dropped again from the GNU accounting utilities. This does not, however, affect the ability to adapt to the acct file format at compile time (when ./configure is run). Even GNU acct 6.3.5 (that does not know about multiple file formats) will yield working binary programs when compiled under a (as yet hypothetical) Linux kernel 2.6.62 that is only able to write the v3 format.

History of the Accounting Utilities

I don't have any idea who originally wrote these utilities. If anybody does, please send some mail to [email protected] and I'll add your information here!

Since the first alpha versions of this software in late 1993, many people have contributed to the package. They are (in alphabetical order):

Eric Backus <[email protected]>
Suggested fixes for HP-UX 9.05 using /bin/cc: configure assumed you were using gcc and tacked on -Wall etc. He also noticed that file_rd.c was doing pointer arithmetic on a void * pointer (non-ANSI).
Christoph Badura <[email protected]>
Christoph was a BIG HELP in computing statistics, most notably k*sec stuff! He also did Xenix testing and contributed some Makefile fixes and output optimizations.
Michael Calwas <[email protected]>
Fixed bugs in mktime.c.
Derek Clegg <[email protected]>
Suggested the simple, elegant fix for *_rd_never_used brain-damage.
Alan Cox <[email protected]>
Original Linux kernel accounting patches.
Scott Crosby <[email protected]>
Suggested idea behind --sort-real-time for sa.
Solar Designer <[email protected]>
Added code for --ahz flag in lastcomm and sa.
Dirk Eddelbuettel <[email protected]>
Managed bug-fixes & etc. for Debian distribution, as well as the architect of merge of GNU + Debian distributions. A big thanks to Dirk for kicking me back into gear again after a long period of no work on this project.
Jason Grant <[email protected]>
Identified a buffer-overrun bug in sa.
Kaveh R. Ghazi <[email protected]>
Tested the package on many systems with compilers other than gcc. Fixed K&R C support.
Susan Kleinmann <[email protected]>
Contributed excellent man pages!
Alexander Kourakos <[email protected]>
Inspired the --wide option for last.
Marek Michalkiewicz <[email protected]>
Suggested the --ip-address flag for last.
David S. Miller <[email protected]>
Noticed missing GNU-standard makefile rules.
Walter Mueller <[email protected]>
Noticed install target was missing, and corrected a typo for prefix in Makefile.in.
Ian Murdock <[email protected]>
Tracked down miscellaneous bugs in sa.c under Linux. Added Debian package maintenance files.
Tuomo Pyhala <[email protected]>
Reported buggy --strict-match flag in lastcomm.
Tim Schmielau <[email protected]>
Added Linux multiformat support.
Luc I. Suryo <[email protected]>
Suggested the --user flag for lastcomm.
Pedro A M Vazquez <[email protected]>
Fixed bugs in sa.c and tested under FreeBSD.
Marco van Wieringen <[email protected]>
Modified (wrote?) Linux kernel accounting patches.


Next: , Previous: Preface, Up: Top

1 ac

The ac command prints out a report of connect time (in hours) based on the logins/logouts in the current wtmp file. A total is also printed out.

The accounting file wtmp is maintained by init and login. Neither of these programs creates the file; if the file is not there, no accounting is done. To begin accounting, create the file with a length of zero. Note that the wtmp file can get really big, really fast. You might want to trim it every once and a while.

GNU ac works nearly the same u*x ac, though it's a little smarter in its printing out of daily totals—it actually prints every day, rather than skipping to the date of the next entry in the wtmp file.

1.1 Flags

All of the original ac's options have been implemented, and a few have been added. Normally, when ac is invoked, the output looks like this:

             total 93867.14

where total is the number of hours of connect time for every entry in the wtmp file. The rest of the flags modify the output in one way or another.

-d
--daily-totals
Print totals for each day rather than just one big total at the end. The output looks like this:
          Jul  3  total     1.17
          Jul  4  total     2.10
          Jul  5  total     8.23
          Jul  6  total     2.10
          Jul  7  total     0.30

-p
--individual-totals
Print time totals for each user in addition to the usual everything-lumped-into-one value. It looks like:
                  bob       8.06
                  goff      0.60
                  maley     7.37
                  root      0.12
                  total    16.15

people
Print out the sum total of the connect time used by all of the users included in people. Note that people is a space separated list of valid user names; wildcards are not allowed.
-f filename
--file filename
Read from the file filename instead of the system's wtmp file.
--complain
When the wtmp file has a problem (a time-warp, missing record, or whatever), print out an appropriate error.
--reboots
Reboot records are not written at the time of a reboot, but when the system restarts; therefore, it is impossible to know exactly when the reboot occurred. Users may have been logged into the system at the time of the reboot, and many ac's automatically count the time between the login and the reboot record against the user (even though all of that time shouldn't be, perhaps, if the system is down for a long time, for instance). If you want to count this time, include the flag. To make ac behave like the one that was distributed with your OS, include this flag.
--supplants
Sometimes a logout record is not written for a specific terminal, so the time that the last user accrued cannot be calculated. If you want to include the time from the user's login to the next login on the terminal (though probably incorrect), include this flag. To make ac behave like the one that was distributed with your OS, include this flag.
--timewarps
Sometimes, entries in a wtmp file will suddenly jump back into the past without a clock change record occurring. It is impossible to know how long a user was logged in when this occurs. If you want to count the time between the login and the time warp against the user, include this flag. To make ac behave like the one that was distributed with your OS, include this flag.
--compatibility
This is shorthand for typing out the three above options.
-a
--all-days
If we're printing daily totals, print a record for every day instead of skipping intervening days where there is no login activity. Without this flag, time accrued during those intervening days gets listed under the next day where there is login activity.
-y
--print-year
Print out the year when displaying dates.
--print-zeros
If a total for any category (save the grand total) is zero, print it. The default is to suppress printing.
--debug
Print verbose internal information.
--tw-leniency value
Set the time warp leniency value (in seconds). Records in wtmp files might be slightly out of order (most notably when two logins occur within a one-second period – the second one gets written first). By default, this value is set to 1 second. Some wtmp's are really screwed up (Suns) and require a larger value here. If the program notices this problem, time is not assigned to users unless the --timewarps flag is used. See the Problems section for more information.
--tw-suspicious value
Set the time warp suspicious value (in seconds). If two records in the wtmp file are farther than this number of seconds apart, there is a problem with the wtmp file (or your machine hasn't been used in a year). If the program notices this problem, time is not assigned to users unless the --timewarps flag is used.
-V
--version
Print ac's version number.
-h
--help
Print ac's usage string and default locations of system files to standard output.

1.2 Problems

For no fault of ac's, if two logins occur at the same time (within a second of each other), each login process will try to write an entry to the wtmp file. With file system overhead, it is forseeable that the entries would get written in the wrong order. GNU ac automatically compensates for this, but some other acs may not... beware.

The FTP Problem

I've tested the standard ac in Ultrix 4.2 (DECstation/DECsystem), SunOS 4.1.1 (Sun3, Sun4, Sparc), Mach 2.5 (Omron/Luna), and DomainOS 10.3 (DN3500). All of these acs have trouble parsing entries in which the line is ftpxxxx (xxxx being some number). Whenever these acs see one of these entries, they log everyone out at the time of the entry.

HOW IT HAPPENS: if there is a user logged into the machine when an ftp connection occurs, (minimally) you'll get a login record for the user, a login record for the ftp connection, and the logouts for both afterwards (in either order).

TANGIBLE RESULT: the user who was logged in gets 'logged out' at the time the ftp connection begins, and none of the time spent during or after the ftp connection. Therefore, when you run GNU ac, the totals will most likely be greater than those of your system's ac (provided you specify the other flags that will make GNU ac behave like the system's).

The Shutdown/Reboot Problem

On Suns, init is a little screwed up. For some reason, after a shutdown record is written, a reboot record is written with a time-stamp before the shutdown (less than 30 seconds, usually).

TANGIBLE RESULT: GNU ac will notice the problem, log everyone out (you can specify if you want the time to be added to the user's total) and begin a new day entry based on the time of the out-of-sync record. If you try to print out daily totals, you'll notice that some days might have two or more entries.

SOLUTION: To fix this, a timewarp leniency value has been implemented. If any record is out of order by this number of seconds (defaults to 60) it gets ignored. If you need to change this value (if you think the totals are off because the value is too high), you can change it using the ‘--timewarp-value’ flag. The rationale for the 60 second default is that of all of the machines with this problem, the largest timewarp was 45.

Stupid System V Machines

Some ac's on System V machines (I've tried SGI Indigo & SGI Indy) forget to pay attention to the ut_type field in a struct utmp. As such, they chalk up a lot of time to non-existant processes called LOGIN or runlevel.

TANGIBLE RESULT: The amount of total time reported by the system's ac is really off. Often, it's several times greater than what it should be.

SOLUTION: GNU ac always pays attention to the ut_type record, so there's no possibility of chalking up time to anything but user processes.


Next: , Previous: ac, Up: Top

2 accton

accton turns process accounting on or off. To save process accounting information in accountingfile, use:

accton accountingfile

2.1 Flags

-V
--version
Print accton's version number.
-h
--help
Print accton's usage string and default locations of system files to standard output.
on
Turns on process accounting using the default accounting file.
off
Turns off process accounting.


Next: , Previous: accton, Up: Top

3 lastcomm

lastcomm prints out information about previously executed commands. If no arguments are specified, lastcomm will print info about all of the commands in the acct file (the record file). If called with a command name, user name, or tty name, only records containing those items will be displayed. For example, to find out which users used command ‘a.out’ and which users were logged into ‘tty0’, type:

lastcomm a.out tty0

This will print any entry for which ‘a.out’ or ‘tty0’ matches in any of the record's fields (command, name, or tty). If you want to find only items that match ALL of the arguments on the command line, you must use the '–strict-match' option. For example, to list all of the executions of command ‘a.out’ by user ‘root’ on terminal ‘tty0’, type:

lastcomm --strict-match a.out root tty0

The order of the arguments is not important.

For each entry the following information is printed:

3.1 Flags

This program implements the features of regular u*x lastcomm with a few extra flags. When lastcomm is invoked without arguments, the output looks like this:

     nslookup         jberman  ttypb      0.03 secs Tue Feb 16 19:23
     comsat           root     __         0.03 secs Tue Feb 16 19:19
     uptime           ctilburg __         0.11 secs Tue Feb 16 19:23
     sh          F    ctilburg __         0.02 secs Tue Feb 16 19:23
     sleep            ctilburg __         0.02 secs Tue Feb 16 19:22
     ls               noel     ttyp4      0.19 secs Tue Feb 16 19:23
--strict-match
Print only entries that match all of the arguments on the command line.
--user name
List records for user with name. This is useful if you're trying to match a username that happens to be the same as a command (e.g., ed).
--command name
List records for command name.
--tty name
List records for tty name.
-f filename
--file filename
Read from the file filename instead of the system's acct file.
--ahz hz
Use this flag to tell the program what AHZ should be (in hertz). This option is useful if you are trying to view an acct file created on another machine which has the same byte order and file format as your current machine, but has a different value for AHZ.
-p
--show-paging
Print paging statistics
--debug
Print verbose internal information.
--version
Print lastcomm's version number.
--help
Print lastcomm's usage string and default locations of system files to standard output.


Next: , Previous: lastcomm, Up: Top

4 sa

sa summarizes information about previously executed commands as recorded in the acct file. In addition, it condenses this data into the savacct summary file, which contains the number of times the command was called and the system resources used. The information can also be summarized on a per-user basis; sa will save this information into usracct. Usage:

sa [opts] [file]

If no arguments are specified, sa will print information about all of the commands in the acct file. If command names have unprintable characters, or are only called once, sa will sort them into a group called ***other. Overall totals for each field are gathered and printed with a blank command name.

If called with a file name as the last argument, sa will use that file instead of acct.

By default, sa will sort the output by sum of user and system time.

The output fields are labeled as follows:

cpu
sum of system and user time in cpu seconds
re
“real time” in cpu seconds
k
cpu-time averaged core usage, in 1k units
avio
average number of I/O operations per execution
tio
total number of I/O operations
k*sec
cpu storage integral (kilo-core seconds)
u
user cpu time in cpu seconds
s
system time in cpu seconds

Note that these column titles do not appear in the first row of the table, but after each numeric entry (as units of measurement) in every row. For example, you might see 79.29re, meaning 79.29 cpu seconds of “real time.”

An asterisk will appear after the name of commands that forked but didn't call exec.

4.1 Flags

The availability of these program options depends on your operating system. In specific, the members that appear in the struct acct of your system's process accounting header file (usually acct.h) determine which flags will be present. For example, if your system's struct acct doesn't have the ac_mem field, the installed version of sa will not support the --sort-cpu-avmem, --sort-ksec, -k, or -K options.

In short, all of these flags may not be available on your machine.

-a
--list-all-names
Force sa not to sort those command names with unprintable characters and those used only once into the `***other' group.
-b
--sort-sys-user-div-calls
Sort the output by the sum of user and system time divided by the number of calls.
-c
--percentages
Print percentages of total time for the command's user, system, and real time values.
-d
--sort-avio
Sort the output by the average number of disk I/O operations.
-D
--sort-tio
Print and sort the output by the total number of disk I/O operations.
-f
--not-interactive
When using the --threshold option, assume that all answers to interactive queries will be affirmative.
-i
--dont-read-summary-file
Don't read the information in savacct.
-j
--print-seconds
Instead of printing total minutes for each category, print seconds per call.
-k
--sort-cpu-avmem
Sort the output by cpu time average memory usage.
-K
--sort-ksec
Print and sort the output by the cpu-storage integral.
-l
--separate-times
Print separate columns for system and user time; usually the two are added together and listed as cpu.
-m
--user-summary
Print the number of processes and number of CPU minutes on a per-user basis.
-n
--sort-num-calls
Sort the output by the number of calls. This is the default sorting method.
-p
--show-paging
Print the number of minor and major pagefaults and swaps.
-P
--show-paging-avg
Print the number of minor and major pagefaults and swaps divided by the number of calls.
-r
--reverse-sort
Sort output items in reverse order.
-s
--merge
Merge the summarized accounting data into the summary files savacct and usracct.
-t
--print-ratio
For each entry, print the ratio of real time to the sum of system and user times. If the sum of system and user times is too small to report—the sum is zero—*ignore* will appear in this field.
-u
--print-users
For each command in the accounting file, print the userid and command name. After printing all entries, quit. Note: this flag supersedes all others.
-v num
--threshold num
Print commands which were executed num times or fewer and await a reply from the terminal. If the response begins with y, add the command to the **junk** group.
--separate-forks
It really doesn't make any sense to me that the stock version of sa separates statistics for a particular executable depending on whether or not that command forked. Therefore, GNU sa lumps this information together unless this option is specified.
--sort-real-time
Sort the output by the “real time” (elapsed time) for each command.
--ahz hz
Use this flag to tell the program what AHZ should be (in hertz). This option is useful if you are trying to view an acct file created on another machine which has the same byte order and file format as your current machine, but has a different value for AHZ.
--debug
Print verbose internal information.
-V
--version
Print sa's version number.
-h
--help
Print sa's usage string and default locations of system files to standard output.

Note: if more than one sorting option is specified, the list will be sorted by the one specified last on the command line.

4.2 Problems

I haven't been able to test this on many different machines because the data files grow so big in a short time; our sysadmin would rather save the disk space.

Most versions of sa that I've tested don't pay attention to flags like --print-seconds and --sort-num-calls when printing out commands when combined with the --user-summary or --print-users flags. GNU sa pays attention to these flags if they are applicable.

4.2.1 mips sa

The average memory use is stored as a short rather than a double, so we suffer from round-off errors. GNU sa uses double the whole way through.


Previous: sa, Up: Top

5 dump-acct

dump-acct dumps some of the contents of one or more acct files in human readable form. Usage:

dump-acct [opts] files

Unless called with the --raw option, it prints a table with the following fields, separated by vertical bars(|):

ac_comm
name of the executed program
ac_version
version of the acct file format
ac_utime
user time
ac_stime
system time
ac_etime
elapsed time
ac_uid
user id
ac_gid
group id
ac_mem
(average) memory usage
ac_io
number of characters transferred on input/output
ac_pid
process id
ac_ppid
parent's process id

All times will be given in platform dependent units (“AHZ”). Not all of the above columns will actually appear, depending on what information your operating system provides in it's struct acct.

5.1 Flags

--ahz hz
Use this flag to tell the program what AHZ should be (in Hertz). This option is useful if you are trying to view an acct file created on another machine which has a different value for AHZ.
--byteswap
Swap the bytes (relative to your system's native byte order) in --raw output.
--format
Set output format with --raw option.
-n num
--num num
Limit the number of lines (or records with --raw) to print.
-r
--reverse
Read the accounting file backwards (print latest record first).
-R
--raw
Don't print human readable output, dump the raw record instead. Useful to convert between different Linux file formats (see below).
-h
--help
Print dump-acct's usage string and default location of the accouning file to standard output.

--byteswap and --format options are only available with Linux multiformat support. They only affect output with the --raw option, format and byte order of the input are automatically detected. Thus they are useful to convert between different file formats.

The --ahz option affects input and output (except for v3 file format, which by definition is fixed to AHZ=100). If you ever need to convert between different AHZ values, use a two-step process: First convert to v3 format with the old AHZ value, then convert to the desired output format with the new AHZ setting.

Table of Contents