sg_disk_io_compare_traffic(3) get disk io statistics

Other Alias

sg_get_disk_io_stats, sg_get_disk_io_stats_r, sg_get_disk_io_stats_diff, sg_get_disk_io_stats_diff_between, sg_free_disk_io_stats, sg_disk_io_compare_name


#include <statgrab.h>

sg_disk_io_stats *sg_get_disk_io_stats (size_t *entries);

sg_disk_io_stats *sg_get_disk_io_stats_r (size_t *entries);

sg_disk_io_stats *sg_get_disk_io_stats_diff (size_t *entries);

sg_disk_io_stats *sg_get_disk_io_stats_diff_between (const sg_disk_io_stats *cur, const sg_disk_io_stats *last, size_t *entries);

sg_error sg_free_disk_io_stats (sg_disk_io_stats *data);

int sg_disk_io_compare_name (const void *va, const void *vb);

int sg_disk_io_compare_traffic (const void *va, const void *vb);


The sg_get_disk_io_stats functions provide disk I/O statistics on a per disk basis. All get- and diff-functions take an optional entries parameter, which points (when given) to a size_t to take the number of returned vector entries.

The sg_get_disk_io_stats() and sg_get_disk_io_stats_r() functions deliver the I/O-statistics since the disk has been attached to the system. The sg_get_disk_io_stats_diff() and sg_get_disk_io_stats_diff_between() deliver the difference between two calls of sg_get_disk_io_stats() or sg_get_disk_io_stats_r(), respectively.

API Shortcut

function returns data owner
sg_get_disk_io_stats sg_disk_io_stats * libstatgrab (thread local)
sg_get_disk_io_stats_r sg_disk_io_stats * caller
sg_get_disk_io_stats_diff sg_disk_io_stats * libstatgrab (thread local)
sg_get_disk_io_stats_diff_between sg_disk_io_stats * caller

sg_disk_io_stats vectors got from sg_get_disk_io_stats_r() or sg_get_disk_io_stats_diff_between() must be freed using sg_free_disk_io_stats() when not needed any more. The caller is responsible for doing it.

Additionally two support functions for qsort(3) are available: sg_disk_io_compare_name() and sg_disk_io_compare_traffic().


size_t entries;
sg_disk_io_stats *io_stats = NULL;
while( NULL != ( io_stats = sg_get_disk_io_stats_diff(&entries) ) ) {
    /* show disks with most traffic first */
    qsort( io_stats, entries, sizeof(io_stats[0]), &sg_disk_io_compare_traffic );
    show_disk_io_stats( io_stats );

On some platforms, such as Solaris 7, the kernel value is stored in a 32bit int, so wraps around when it reaches 4GB. Other platforms, such as Solaris 8 (and most other modern systems), hold the value in a 64bit int, which wraps somewhere near 17 million terabytes. The sg_get_disk_io_stats_diff() function and the sg_get_disk_io_stats_diff_between() function care about these overflows and try to detect overflows when the diff is calculated.

On Solaris libstatgrab will attempt to get the cXtXdXsX representation for the disk_name string. If it fails it will use a name like sd0. On some systems programs calling libstatgrab will need elevated privileges to lookup some of the names. The mappings are built up when sg_init() is called for the first time.


All diskio statistics return a pointer to a structure of type sg_disk_io_stats.

typedef struct {
        char *disk_name;
        unsigned long long read_bytes;
        unsigned long long write_bytes;
        time_t systime;
} sg_disk_io_stats;
The name known to the operating system. (eg. on linux it might be hda)
The number of bytes the disk has read.
The number of bytes the disk has written.
The time period over which read_bytes and write_bytes were transferred.


sg_get_disk_io_stats_diff and sg_get_disk_io_stats_diff_between compare two lists of disk (block device) related I/O statistics. Each entry occurring only in the second list is passed through to the resulting list as if it would have been compared to an entry with all statistic values set to 0. This implies, on the very first call sg_get_disk_io_stats_diff will return the same as sg_get_disk_io_stats.

On operating systems that hold only 32bits of data there is a problem if the values wrap twice. For example, on Solaris 7 if 9GB is transferred and the operating system wraps at 4GB, the sg_get_disk_io_stats_diff() function will return 5GB.

The compare functions exists rather for backward compatibility than for functionality enhancements. Limited flexibility (e.g. reverse order) and lack of optimising opportunities for the compiler leads to the recommendation to implement the required compare routines locally.