MONITOR(1L) MONITOR(1L) NAME monitor - display and update information about the AIX/6000 system events and optionally top cpu processes SYNOPSIS monitor [-options] DESCRIPTION Monitor shows various system variables and updates the screen periodically. Monitor can also be used to show top cpu processes. The metrics can also be logged into a file to build historical performance information of your system. Monitor can be exited by typing 'q' or by typing ctrl-C (keyboard interrupt). The running sample can be ended by pressing any key. Typing ctrl-L ends the sample but also redraws the entire screen. Other interactive key commands are a to toggle between normal and alternative display. d to toggle full disk disk display in alternative display. n to toggle full net display in alternative display. N to toggle NFS display (mutually exclusive with AFS). B to toggle AFS display (mutually exclusive with NFS). s to toggle smp cpu display in alternate display. t to toggle top processes display. u to toggle between username and userid in top. r to toggle between showing running or all processes. c, p, m, v to sort top processes on cpu usage, page faults, main memory usage or virtual memory usage. R to toggle includeing root sessions in user count/activity. +, - to increase or decrease to sample time by one second. >, < to increase or decrease to sample time by five seconds. Page 1 (printed 6/23/01) MONITOR(1L) MONITOR(1L) ?, h show this help message. All commands work in both normal and alternative display mode, but for example the d and n commands do not automatically switch between these display modes. The 't' command is special: it remembers top display status for both display modes seperately (although the other top related commands operate on both display modes at the same time.) Note: changing the sample time may influence the Sync option. When logmode is also on, sample time will never exceed the interval time. The first two samples after a sample time change may seem to have strange sample times. This is normal. Note: Monitor has support for AFS, but this only works when you have AFS installed. AIX systems before version 3.2.4 do not calculate load averages. Monitor uses another process to calculate load average values in background. This process is started automatically first time some user tries to use this monitor or any other program which tries to get loadavgvalues with supplied getloadavg-subroutine. There are two output modes for monitor: on screen or to a file. Both can be active at the same time. Normally you use the screen output for real-time monitoring or ad hoc troubleshooting. The file output mode can be used for general performance monitoring and capacity planning. When needed, both can be active at the same time. OPTIONS Note: all options and option arguments can be abbreviated to their first character except for the -smp option (as it overloads with -sample). Options are given in alphabetical (ignoring case) order. -alternative Shows both system events and top cpu processes. Note this will need higher window (eg. Xwindows xterm or aixterm). -Active_threshold seconds To calculate user activity, the terminal is checked on last modification date. Terminals modified longer than seconds ago are counted as an inactive user. Default is 60 seconds. Page 2 (printed 6/23/01) MONITOR(1L) MONITOR(1L) -B Show AFS information when AFS is running. This option is mutually exclusive with the -Nfs option. This only works if you have AFS, and monitor is compiled with AFS enabled. -compress [prog] When the filename for logmode changes, this option will compress the old logfile using prog as pipe. Default for prog is /usr/local/bin/gzip. Note, compression only works for root or users in the system group, to avoid security leaks. -Count count Display/log only count screen refreshes/samples (Note: this counts samples not intervals!) -disk Show detailed disk information. -Disk [count] Sort disk information on read+write rate. When count is specified, only that number of disks are shown. -Filesys Also log non-local filesystem information like NFS and AFS. Beware: This will request information from the NFS/AFS after every interval. -Highlight [none|reverse|bold|c#|m#|v#] Set highlighting mode of headers. Default is set during compilation, see monitor -h for your setting. c can be used to set a header color if your terminal supports color. m can be used to specify metric name color if your terminal supports color. v can be used to specify metric value color if your terminal supports color. For color terminals an array is filled with foreground/background pairs from which a setting can be chosen. If your system supports 8 colors, the first 8 colors (starting from 0) have backgroundcolor 0, the second 8 have backgroundcolor 1 etc. -interval time Set interval length to time seconds for logmode. Default 900 seconds. Page 3 (printed 6/23/01) MONITOR(1L) MONITOR(1L) -Interactive Enter screen mode. When the -log option is used, the - Interactive option is needed to do screen mode simultaneously. Otherwise this is default. -log logfile Switch on log mode. This will dump the measured data on interval basis to logfile. The Logfile can be specified as a strftime(3) format string, based on start of interval time. The logfile will be opened/closed for every dump. So when time passes, the logfilename may change depending on the strftime format. (see also -compress option). Hint: The interval record in the logfile contains the weekday number with sunday being the first day of the week (day 0). When logs are created on a weekly basis (I do), use %U instead of %W so logfiles are switched every sunday. Logmode implies synchronized intervals (see -Sync option). During an interval, several samples will be taken (see -sample option). In logging mode, the maximum values of the sample data during the interval are also written to the logfile. -L Dump samples data to the logfile if logmode is on. -net Show detailed network statistics. -Nfs Show nfs information when nfs is running. When AFS support is available this option is mutually exclusive with the -B option. -run Show only running processes or show those processes which have gained CPU time. -Rootinclude By default root sessions are ignored in user count/activity calculations. This option will include them. -sample time Set the delay between screen updates to time seconds. The default delay is 10 seconds after first initial 1 to 2 second delay (see also -Sync option). Note: smaller sampletimes take more cputime. Page 4 (printed 6/23/01) MONITOR(1L) MONITOR(1L) -smp Show symmetric multiprocessing cpu information. -Sync Synchronise sample and interval time on day boundary. This means that when you specify an interval time of 15 minutes (-i 900), the intervals will end at for 0:00, 0:15, 0:30, 0:45, 1:00 etc. This option is implied by the -log option. -top [nproc] Shows top cpu processes and shows only summary of system variables. If extra parameter is given monitor will show that amount of processes. If display is wider than 94 characters, top-display will show how user and system time distribute among processes. It also shows process status in terms of run sleep etc. In fromt of thes status indicator you my find an a terminal. Without the 'F' the process is backgrounded. -Toplog When log mode is on, also log top records to logfile. -user Show only uid from users. This could make the top display faster if your passwd database is not fast enough. (Eg. you have more than 2000 passwd entries and you are not using passwd.pag/.dir databases made with mkpasswd-command). DISPLAY In default mode monitor will show the following information o Hostname, date and time between display delay. o Percentage distribution of cpu-load (system, wait, user and idle percentages) with graph where '=' means system, 'W' means wait, ">" means user and "." means idle time. o Runnable processes value/second and load average values of 1, 5 and 15 minute times. o Processes waiting to be swapped in. o Both free and total from real and virtual memory, real memory is RAM-memory and virtual memory is paging space. For real memory the usage is shown for process Page 5 (printed 6/23/01) MONITOR(1L) MONITOR(1L) and file pages. In AIX normal file accesses are done internally by mapping files to memory and thus memory seems to be always used. o Paging information; pagefaults, pages to be paged in and out from user space and pages to be paged in and out from paging (swap) space, page size in AIX/6000 is 4 kB. o Various process and system events: process (context) switches, system call, read and write -calls, forks and execs, three interrupts rcvint, xmtint, mdmint. o File and tty-IO variables; iget, namei and dirblk, amount of read and written kbytes, amount of tty-based characters handled. o Disk-information, read and written bytes/second and amount of busy time. o NFS RPC information of read,write and other access calls. o Network (netw) information; amount of read and written kB per second. When using -top option top lines will show: o Hostname, load average values of 1, 5 and 15 minutes and date. o Percentage distribution of cpu states (user, system, wait and idle time). o For real and virtual memory the amount used, free and total memory. Following lines will show values of the top cpu time processes sorted by current cpu-usage as first sortkey and by second key the total amount cputime used. The display of processes is similar to ps(1). PID is the process id. USER is the username of the process owner. PRI is the current priority of the process. NICE is the nice value for the process in the range -20 to 20 (note for kernel-processes the nice value is 21 which means that the process has fixed-priority and the nice value doesn't have any meaning). SIZE is the amount virtual memory used. RES is the amount of real memory used. STAT is the status of the process, a "F" in front of the STAT value indicates a foreground job. TIME is the amount of cputime used. When your screen is wide enough, the cputime is split in Page 6 (printed 6/23/01) MONITOR(1L) MONITOR(1L) USER/SYSTIME: User mode processing and kernel mode processing. CPU% comes in two parts. The left part indicates the cputime distribution withing the interval and the right part is the cpu usage since startup of that process. COMMAND is the short-name of the process command. If process was kernel process "Kernel" is added to process name. Monitor -top will display both Kernel and user processes in process list. Kernel processes are displayed like which consumes all idle CPU cycles. See FAQ list for more types of kernel processes. On wider screens (more than 94 columns) the top display shows the user and system time distribution for each process. SECURITY To allow any user to run monitor it has to be setuid root or setgid system. This is needed to access kernel information. As soon as the access pattern for this data is setup, root and system permissions are given up (this happens within the first few milliseconds). Only after giving up, logfiles may be opened and the compress program may be run. As a consequence, logfiles can only be created where the real user has permissions and ownership will be set to the real user! VERSION This manual page is for monitor version 2.1.5. AUTHOR Marcel Mol, MESA Consulting, The Netherlands Email: marcel@mesa.nl URL: http://www.mesa.nl/~marcel http://www.mesa.nl/pub/monitor ftp://ftp.mesa.nl/pub/monitor Jussi Maki, Center for Scientific Computing, Finland Email: jmaki@csc.fi URL: http://www.csc.fi/~jmaki SEE ALSO vmstat(1), ps(1), sar(1), iostat(1), nfsstat(1) Page 7 (printed 6/23/01)