Get information about resource utilization
#include <sys/resource.h> int getrusage( int who, struct rusage * r_usage );
- Which process to get the usage for:
- RUSAGE_CHILDREN — get information about resources used by the terminated and waited-for children of the current process. If the child is never waited for (e.g if the parent has SA_NOCLDWAIT set, or sets SIGCHLD to SIG_IGN), the resource information for the child process is discarded and isn't included.
- RUSAGE_SELF — get information about resources used by the current process.
- A pointer to an object of type struct rusage in which the function can store the resource information; see below.
Use the -l c option to qcc to link against this library. This library is usually included automatically.
The getrusage() function provides measures of the resources used by the current process or its terminated and waited-for child processes, depending on the value of the who argument.
The rusage structure is defined as:
struct timeval ru_utime; /* user time used */ struct timeval ru_stime; /* system time used */ long ru_maxrss; /* max resident set size */ long ru_ixrss; /* integral shared memory size */ long ru_idrss; /* integral unshared data " */ long ru_isrss; /* integral unshared stack " */ long ru_minflt; /* page reclaims */ long ru_majflt; /* page faults */ long ru_nswap; /* swaps */ long ru_inblock; /* block input operations */ long ru_oublock; /* block output operations */ long ru_msgsnd; /* messages sent */ long ru_msgrcv; /* messages received */ long ru_nsignals; /* signals received */ long ru_nvcsw; /* voluntary context switches */ long ru_nivcsw; /* involuntary " */
The members include:
- The total amount of time, in seconds and microseconds, spent executing in user mode.
- The total amount of time, in seconds and microseconds, spent executing in system mode.
- The maximum resident set size, given in pages. See the Caveats section, below.
- Not currently supported.
- An integral value indicating the amount of memory in use by a process while the process is running. This value is the sum of the resident set sizes of the process running when a clock tick occurs. The value is given in pages times clock ticks. It doesn't take sharing into account. See the Caveats section, below.
- Not currently supported.
- The number of page faults serviced that didn't require any physical I/O activity. See the Caveats section, below.
- The number of page faults serviced that required physical I/O activity. This could include page ahead operations by the kernel. See the Caveats section, below
- The number of times a process was swapped out of main memory.
- The number of times the file system had to perform input in servicing a read() request.
- The number of times the filesystem had to perform output in servicing a write() request.
- The number of messages sent over sockets.
- The number of messages received from sockets.
- The number of signals delivered.
- The number of times a context switch resulted due to a process's voluntarily giving up the processor before its timeslice was completed (usually to await availability of a resource).
- The number of times a context switch resulted due to a higher priority process's becoming runnable or because the current process exceeded its time slice.
- An error occurred (errno is set).
- The address specified by the r_usage argument isn't in a valid portion of the process's address space.
- Invalid who parameter.
Only the timeval fields of struct rusage are supported.
The numbers ru_inblock and ru_oublock account only for real I/O, and are approximate measures at best. Data supplied by the cache mechanism is charged only to the first process to read and the last process to write the data.
The way resident set size is calculated is an approximation, and could misrepresent the true resident set size.
Page faults can be generated from a variety of sources and for a variety of reasons. The customary cause for a page fault is a direct reference by the program to a page that isn't in memory. Now, however, the kernel can generate page faults on behalf of the user, for example, servicing read() and write() functions. Also, a page fault can be caused by an absent hardware translation to a page, even though the page is in physical memory.
In addition to hardware-detected page faults, the kernel may cause pseudo page faults in order to perform some housekeeping. For example, the kernel may generate page faults, even if the pages exist in physical memory, in order to lock down pages involved in a raw I/O request.
By definition, major page faults require physical I/O, while minor page faults don't. For example, reclaiming the page from the free list would avoid I/O and generate a minor page fault. More commonly, minor page faults occur during process startup as references to pages which are already in memory. For example, if an address space faults on some hot executable or shared library, a minor page fault results for the address space. Also, anyone doing a read() or write() to something that's in the page cache gets a minor page fault(s) as well.
There's no way to obtain information about a child process that hasn't yet terminated.