8 With the introduction of cfq v3 (aka cfq-ts or time sliced cfq), basic io
9 priorities are supported for reads on files. This enables users to io nice
10 processes or process groups, similar to what has been possible with cpu
11 scheduling for ages. This document mainly details the current possibilities
12 with cfq; other io schedulers do not support io priorities thus far.
17 CFQ implements three generic scheduling classes that determine how io is
20 IOPRIO_CLASS_RT: This is the realtime io class. This scheduling class is given
21 higher priority than any other in the system, processes from this class are
22 given first access to the disk every time. Thus it needs to be used with some
23 care, one io RT process can starve the entire system. Within the RT class,
24 there are 8 levels of class data that determine exactly how much time this
25 process needs the disk for on each service. In the future this might change
26 to be more directly mappable to performance, by passing in a wanted data
29 IOPRIO_CLASS_BE: This is the best-effort scheduling class, which is the default
30 for any process that hasn't set a specific io priority. The class data
31 determines how much io bandwidth the process will get, it's directly mappable
32 to the cpu nice levels just more coarsely implemented. 0 is the highest
33 BE prio level, 7 is the lowest. The mapping between cpu nice level and io
34 nice level is determined as: io_nice = (cpu_nice + 20) / 5.
36 IOPRIO_CLASS_IDLE: This is the idle scheduling class, processes running at this
37 level only get io time when no one else needs the disk. The idle class has no
38 class data, since it doesn't really apply here.
43 See below for a sample ionice tool. Usage:
45 # ionice -c<class> -n<level> -p<pid>
47 If pid isn't given, the current process is assumed. IO priority settings
48 are inherited on fork, so you can use ionice to start the process at a given
51 # ionice -c2 -n0 /bin/ls
53 will run ls at the best-effort scheduling class at the highest priority.
54 For a running process, you can give the pid instead:
56 # ionice -c1 -n2 -p100
58 will change pid 100 to run at the realtime scheduling class, at priority 2.
60 ---> snip ionice.c tool <---
67 #include <sys/ptrace.h>
68 #include <asm/unistd.h>
70 extern int sys_ioprio_set(int, int, int);
71 extern int sys_ioprio_get(int, int);
74 #define __NR_ioprio_set 289
75 #define __NR_ioprio_get 290
76 #elif defined(__ppc__)
77 #define __NR_ioprio_set 273
78 #define __NR_ioprio_get 274
79 #elif defined(__x86_64__)
80 #define __NR_ioprio_set 251
81 #define __NR_ioprio_get 252
82 #elif defined(__ia64__)
83 #define __NR_ioprio_set 1274
84 #define __NR_ioprio_get 1275
86 #error "Unsupported arch"
89 static inline int ioprio_set(int which, int who, int ioprio)
91 return syscall(__NR_ioprio_set, which, who, ioprio);
94 static inline int ioprio_get(int which, int who)
96 return syscall(__NR_ioprio_get, which, who);
107 IOPRIO_WHO_PROCESS = 1,
112 #define IOPRIO_CLASS_SHIFT 13
114 const char *to_prio[] = { "none", "realtime", "best-effort", "idle", };
116 int main(int argc, char *argv[])
118 int ioprio = 4, set = 0, ioprio_class = IOPRIO_CLASS_BE;
121 while ((c = getopt(argc, argv, "+n:c:p:")) != EOF) {
124 ioprio = strtol(optarg, NULL, 10);
128 ioprio_class = strtol(optarg, NULL, 10);
132 pid = strtol(optarg, NULL, 10);
137 switch (ioprio_class) {
138 case IOPRIO_CLASS_NONE:
139 ioprio_class = IOPRIO_CLASS_BE;
141 case IOPRIO_CLASS_RT:
142 case IOPRIO_CLASS_BE:
144 case IOPRIO_CLASS_IDLE:
148 printf("bad prio class %d\n", ioprio_class);
153 if (!pid && argv[optind])
154 pid = strtol(argv[optind], NULL, 10);
156 ioprio = ioprio_get(IOPRIO_WHO_PROCESS, pid);
158 printf("pid=%d, %d\n", pid, ioprio);
161 perror("ioprio_get");
163 ioprio_class = ioprio >> IOPRIO_CLASS_SHIFT;
164 ioprio = ioprio & 0xff;
165 printf("%s: prio %d\n", to_prio[ioprio_class], ioprio);
168 if (ioprio_set(IOPRIO_WHO_PROCESS, pid, ioprio | ioprio_class << IOPRIO_CLASS_SHIFT) == -1) {
169 perror("ioprio_set");
174 execvp(argv[optind], &argv[optind]);
180 ---> snip ionice.c tool <---
183 March 11 2005, Jens Axboe <jens.axboe@oracle.com>