prio(5)prio(5)NAMEprio - priority I/O
DESCRIPTION
Priority I/O (prio) refers to adding a "quality of service" to file
descriptors in the form of hardware bandwidth allocation and priority
request scheduling. The purpose of this mechanism is to manage the
sharing of scarce IO resources amongst competing processes, and to permit
a given process to reserve a portion of the system's resources for its
exclusive use for a period of time.
Bandwidth allocation is operated on a file descriptor. Currently, the
file descriptor must point to a regular file. In the future more file
types may be added. The data flow is assumed to be between memory and
the hardware device. When an application makes a bandwidth allocation
request, each hardware component along the path will try to allocate
required bandwidth. Any failure to allocate along the path causes the
entire allocation to fail. Given the limited nature of the bandwidth
resource, read and write are separate allocations.
The sequence you use the API calls are: 1) if you want to do a series of
allocation and reallocation and don't want other applications to
interfere, do prioLock(); 2) request bandwidth allocation by doing
prioSetBandwidth(); 3) if you want to get bandwidth allocation
information, do prioGetBandwidth(); 4) if you did prioLock(), you must do
prioUnlock() to release the lock so that other applications can proceed
to get their bandwidth allocation. If an application forgets to release
the lock, the system will release the lock when the application
terminates.
A regular file must be on a XFS file system, and do DIRECT I/O (see
fcntl(5)) in order to use priority IO functionalities. If the file system
resides on a logical volume, it must be striped across all the disks in
the volume. On systems that don't have the bandwidth allocation hardware
, only priority scheduling is done. The "priority" attribute obeys the
normal file attribute sharing semantics.
Some amount of system configuration needs to be done before "priority
I/O" can be used on the file system. By default, the disk bandwidth is
assumed to be zero. A utility disk_bandwidth (see disk_bandwidth(1)) has
been provided which can be used to measure the disk bandwidth. This must
be run on each disk on which "priority I/O" reservations are expected to
be made. After running disk_bandwidth also run ioconfig (1) in order to
avail of the just measured bandwidths. See disk_bandwidth(1) for some
cautionary notes before using the utility.
EXAMPLE
char *filename;
int fd, holder;
fd = open(filename, O_RDONLY);
if (prioSetBandwidth(fd, PRIO_READ_ALLOCATE, 200000, &holder) < 0)
Page 1
prio(5)prio(5)
if (errno == ENOLCK) {
printf("The lock holder is: %d\n", holder);
exit(1);
}
NOTE
In IRIX 6.5, the Priority I/O APIs have been merged with the GRIO APIs.
As a result, the use of the Priority I/O APIs is deprecated. For
backwards compatibility, prioSetBandwidth and prioGetBandwidth are still
supported. prioLock and prioUnlock are merely stubs. No locking or
atomicity is guaranteed. Please refer to grio(5) for further details.
SEE ALSOdisk_bandwidth(1m), prio(3X), grio(5).
Page 2
