Operating SystemsCPU Scheduling

Priority Scheduling

LevelIntermediate

Duration90 mins

TopicCPU Scheduling

1 / 5

Priority Assignment

The Priority Imperative in Modern Computing

In the complex orchestration of modern operating systems, not all processes are created equal. A nuclear power plant's safety monitoring system cannot wait in the same queue as a background file indexer. A real-time video conferencing application demands different treatment than a nightly backup job. The mechanism that enables this differentiation—priority assignment—forms the foundation of intelligent CPU scheduling.

Priority assignment is far more than a simple numbering scheme. It represents a sophisticated contract between the operating system and its processes, encoding expectations about urgency, importance, resource requirements, and temporal constraints. Understanding how priorities are assigned, interpreted, and managed is essential for anyone seeking to comprehend how operating systems balance the competing demands of dozens or hundreds of concurrent processes.

What You Will Learn

By the end of this page, you will understand the fundamental principles of priority assignment, including static versus dynamic priorities, internal versus external priority sources, the mathematical foundations of priority calculation, and how different operating systems implement priority schemes. You will be equipped to analyze and predict scheduling behavior based on priority configurations.

Foundations of Priority in Operating Systems

Priority, in the context of CPU scheduling, is a quantitative or qualitative measure that determines the relative urgency with which a process should receive CPU time. This seemingly simple definition conceals layers of complexity that operating system designers must navigate.

The Semantic Challenge of Priority

Before examining mechanisms, we must address a fundamental semantic ambiguity: what does a priority number actually mean? Different operating systems adopt diametrically opposite conventions:

Lower number = Higher priority: Used in UNIX-like systems (Linux, macOS, BSD). A process with priority 0 is more important than one with priority 10.
Higher number = Higher priority: Used in Windows and many real-time operating systems. A thread with priority 31 runs before one with priority 15.

This inconsistency has caused countless bugs when developers move between platforms. The conceptual priority (importance) maps to numerical priority differently depending on the operating system.

Priority Convention Comparison Across Operating Systems
Operating System	Priority Range	Convention	Highest Priority Value	Lowest Priority Value
Linux (Nice values)	-20 to +19	Lower = Higher	-20	+19
Linux (Real-time)	0 to 99	Higher = Higher	99	0
Windows	0 to 31	Higher = Higher	31 (Real-time)	0 (Idle)
macOS (Mach)	0 to 127	Lower = Higher	0	127
FreeBSD	-20 to +20	Lower = Higher	-20	+20
VxWorks (RTOS)	0 to 255	Lower = Higher	0	255
FreeRTOS	0 to configMAX_PRIORITIES-1	Higher = Higher	configMAX_PRIORITIES-1	0

Priority Inversion of Semantics

When writing cross-platform code or analyzing system behavior, always verify the priority convention. A priority value of '1' means vastly different things on Linux versus Windows. Documentation that simply states 'set priority to 10' without context is dangerously ambiguous.

Priority as a Scheduling Parameter

Mathematically, priority functions as a partial ordering over the set of ready processes. Given processes P₁, P₂, ..., Pₙ in the ready queue, the scheduler selects process Pᵢ such that:

priority(Pᵢ) ≥ priority(Pⱼ) for all j ≠ i

Or, in systems where lower values indicate higher priority:

priority(Pᵢ) ≤ priority(Pⱼ) for all j ≠ i

This ordering determines CPU allocation decisions, but the derivation of priority values—how they are assigned and updated—varies dramatically across scheduling implementations.

Static Priority Assignment

Static priority (also called fixed priority) assigns a priority value to a process at creation time, which remains constant throughout the process's lifetime. This approach offers predictability and simplicity but sacrifices adaptability to changing system conditions.

Characteristics of Static Priority

Static priority systems exhibit several defining characteristics:

Determinism: Given initial priority assignments, the scheduling sequence is fully predictable. This property is crucial for real-time systems where timing guarantees must be made.
Simplicity: No runtime computation is required to update priorities. The scheduler simply compares fixed values.
Stability: Process behavior cannot influence its scheduling priority. A process cannot 'game' the system by altering its execution patterns.
Potential for Starvation: Low-priority processes may never execute if higher-priority processes remain runnable indefinitely.

Common Static Priority Assignment Strategies

•Process Type Classification — System processes receive higher priority than user processes. Kernel threads outrank application threads. Device drivers may occupy intermediate levels.
•User Specification — Unix nice values allow users to voluntarily lower their process priority (or raise it, if privileged). Windows allows priority class specification via SetPriorityClass().
•Application-Defined Priority — Software architects designate priority levels during design. A web server's request handler might run at higher priority than its logging subsystem.
•Resource Ownership — Processes holding critical resources (locks, device handles) may receive elevated priority to release those resources quickly.
•Rate Monotonic Assignment — In real-time systems, processes with shorter periods receive higher priority. This is provably optimal for certain task sets.

static_priority_examples.c
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
#include <sched.h>
#include <pthread.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <errno.h>
 
/**
 * Demonstrates static priority assignment using POSIX real-time scheduling.
 * Real-time priorities in Linux range from 1 to 99, with 99 being highest.
 * Requires CAP_SYS_NICE capability or root privileges.
 */
 
void* high_priority_task(void* arg) {
    printf("High priority task executing (priority 80)\n");
    // Critical real-time work here
    volatile long sum = 0;
    for (long i = 0; i < 100000000; i++) {
        sum += i;
    }
    printf("High priority task completed\n");
    return NULL;
}
 
void* low_priority_task(void* arg) {
    printf("Low priority task executing (priority 20)\n");
    // Less critical work here
    volatile long sum = 0;
    for (long i = 0; i < 100000000; i++) {
        sum += i;
    }
    printf("Low priority task completed\n");
    return NULL;
}
 
int main() {
    pthread_t high_thread, low_thread;
    pthread_attr_t high_attr, low_attr;
    struct sched_param high_param, low_param;
    
    // Initialize thread attributes
    pthread_attr_init(&high_attr);
    pthread_attr_init(&low_attr);
    
    // Set scheduling policy to SCHED_FIFO (real-time, static priority)
    pthread_attr_setschedpolicy(&high_attr, SCHED_FIFO);
    pthread_attr_setschedpolicy(&low_attr, SCHED_FIFO);
    
    // Enable explicit scheduling (don't inherit from parent)
    pthread_attr_setinheritsched(&high_attr, PTHREAD_EXPLICIT_SCHED);
    pthread_attr_setinheritsched(&low_attr, PTHREAD_EXPLICIT_SCHED);
    
    // Assign static priorities
    high_param.sched_priority = 80;  // High priority (1-99 range)
    low_param.sched_priority = 20;   // Low priority
    
    pthread_attr_setschedparam(&high_attr, &high_param);
    pthread_attr_setschedparam(&low_attr, &low_param);
    
    // Create threads with static priorities
    // Note: Low priority thread created first, but high priority runs first
    if (pthread_create(&low_thread, &low_attr, low_priority_task, NULL) != 0) {
        perror("Failed to create low priority thread");
        fprintf(stderr, "Error: %s\n", strerror(errno));
        fprintf(stderr, "Hint: Run with sudo or set CAP_SYS_NICE\n");
        exit(1);
    }
    
    if (pthread_create(&high_thread, &high_attr, high_priority_task, NULL) != 0) {
        perror("Failed to create high priority thread");
        exit(1);
    }
    
    // Wait for completion
    pthread_join(high_thread, NULL);
    pthread_join(low_thread, NULL);
    
    // Clean up
    pthread_attr_destroy(&high_attr);
    pthread_attr_destroy(&low_attr);
    
    printf("Both tasks completed\n");
    return 0;
}

Rate Monotonic Scheduling (RMS)

In real-time systems, Rate Monotonic Scheduling assigns static priorities based on task periods: shorter period → higher priority. Liu and Layland proved (1973) that RMS is optimal among fixed-priority algorithms for independent, periodic tasks on a single processor. If a task set is schedulable by any fixed-priority algorithm, RMS will schedule it.

Dynamic Priority Assignment

Dynamic priority adjusts a process's priority during execution based on various runtime factors. This approach sacrifices predictability for adaptability, allowing the scheduler to respond to changing system conditions and process behaviors.

The Need for Dynamic Priorities

Static priorities, while simple, cannot account for:

Behavioral Changes: A process may alternate between CPU-intensive and I/O-intensive phases. Optimal priority differs between phases.
System Load Variations: The 'right' priority for a process depends on what else is running. A background task might safely run at higher priority when the system is idle.
Fairness Requirements: Without dynamic adjustment, high-priority processes can monopolize the CPU, starving lower-priority work.
Responsiveness Demands: Interactive processes need priority boosts when users interact with them, even if they previously ran at lower priority.

Dynamic Priority Computation

Dynamic priorities are typically computed as a function of multiple factors:

dynamic_priority_formula.txt

Mathematical Model

Dynamic Priority Computation (General Model)
═══════════════════════════════════════════════════════════════════════
 
    effective_priority(P, t) = base_priority(P) 
                              + f(cpu_usage(P, t)) 
                              + g(wait_time(P, t))
                              + h(io_behavior(P, t))
                              + bonus(P, t)
 
Where:
────────────────────────────────────────────────────────────────────────
• base_priority(P)     : Initial priority (may be user-specified)
• cpu_usage(P, t)      : Recent CPU consumption (typically decreases priority)
• wait_time(P, t)      : Time spent waiting (may increase priority for fairness)
• io_behavior(P, t)    : I/O-bound processes often receive priority boosts
• bonus(P, t)          : Context-dependent adjustments (interactive use, etc.)
 
Linux CFS Example (Simplified Virtual Runtime):
────────────────────────────────────────────────────────────────────────
    vruntime(P) = vruntime(P) + (delta_exec * NICE_0_LOAD) / weight(P)
    
    Where weight(P) is derived from nice value:
        weight(nice) = 1024 / (1.25 ^ nice)
    
    Process with LOWEST vruntime runs next (most "starved" of CPU time).
 
Windows Priority Boost Example:
────────────────────────────────────────────────────────────────────────
    When I/O completes:
        boosted_priority = base_priority + io_boost
        (io_boost depends on device type: keyboard > disk > network)
    
    Boost decays: After each quantum without blocking:
        priority = max(priority - 1, base_priority)
 
FreeBSD ULE Scheduler:
────────────────────────────────────────────────────────────────────────
    priority = PRI_MIN_TIMESHARE 
               + (recent_cpu_ticks * INVERSE_ESTCPU_WEIGHT)
               - (interactivity_score * INVERSE_INTERACT_WEIGHT)
    
    Interactivity score rewards processes that sleep frequently.

Factors Influencing Dynamic Priority

•CPU Consumption History — Processes that have consumed significant CPU time recently see priority reductions. This prevents CPU-bound processes from monopolizing the processor.
•Time Spent Waiting — Processes waiting in the ready queue may receive priority increases (aging). This prevents indefinite starvation of low-priority processes.
•I/O Behavior Patterns — I/O-bound processes typically receive priority boosts. They voluntarily yield the CPU while waiting for I/O, so boosting their priority when ready improves overall system throughput.
•Interactive Events — Keyboard input, mouse clicks, and window focus changes may trigger priority boosts for the associated process, improving perceived responsiveness.
•Real-time Constraints — Approaching deadlines may increase priority in deadline-aware schedulers. Missing a deadline may decrease priority or trigger rescheduling.
•Resource Contention — Processes holding locks or other synchronization primitives may receive temporary boosts to release those resources quickly, reducing overall blocking time.

Decay Functions and Temporal Relevance

Dynamic priority systems typically weight recent behavior more heavily than past behavior. Exponential decay functions (e.g., weighted averages with decay factor α) ensure that a process's current priority reflects its recent activity rather than ancient history. A process that was CPU-intensive an hour ago but has been idle for 59 minutes shouldn't be penalized for that past behavior.

Internal vs External Priority Sources

Priority assignment sources can be categorized as internal (determined by the operating system based on observable process characteristics) or external (specified by users, administrators, or applications). Most modern schedulers blend both sources.

Internal Priority Factors

Internal factors are computed automatically by the operating system:

Internal Priority Factors
Factor	Measurement	Typical Effect	Rationale
CPU Burst Length	Runtime between I/O operations	Shorter bursts → higher priority	Favor I/O-bound processes for throughput
I/O Wait Ratio	Time waiting for I/O / total time	Higher ratio → higher priority	I/O-bound processes release CPU voluntarily
Memory Usage	Working set size, page fault rate	Varies by policy	May penalize memory-hungry processes
Time in System	Total elapsed time since creation	Longer → higher priority (aging)	Prevent indefinite starvation
Sleep Frequency	How often process voluntarily yields	More sleep → higher priority	Reward cooperative behavior
Recent CPU Time	CPU seconds in last N milliseconds	More CPU → lower priority	Share CPU fairly among processes

External Priority Factors

External factors come from outside the kernel's observations:

External Priority Factors
Factor	Source	Mechanism	Example
User-specified Nice Value	User or script	`nice`, `renice` commands	`nice -n 10 ./background_job`
Process Priority Class	Application API	System call at creation	`SetPriorityClass()` on Windows
Real-time Classification	Administrator/application	Scheduler policy selection	`sched_setscheduler()` with `SCHED_FIFO`
Control Group (cgroup) Weights	System administrator	cgroup configuration files	Set cpu.weight in cgroup v2
Container/VM Resource Limits	Orchestration system	Container runtime configuration	Kubernetes resource requests/limits
Affinity Settings	Application or admin	Processor affinity masks	`taskset` command or `pthread_setaffinity_np()`

priority_sources.sh
Shell
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
#!/bin/bash
# Demonstrating various external priority mechanisms on Linux
 
# ═══════════════════════════════════════════════════════════════
# Nice Values: Traditional UNIX priority adjustment
# ═══════════════════════════════════════════════════════════════
 
# Start a process with reduced priority (nice value +10)
nice -n 10 ./cpu_intensive_job &
NICE_PID=$!
echo "Started CPU job with nice 10 (PID: $NICE_PID)"
 
# Adjust priority of running process
renice +15 -p $NICE_PID
echo "Adjusted nice value to 15"
 
# View nice values of all user processes
ps -eo pid,ni,comm --sort=-ni | head -20
 
# ═══════════════════════════════════════════════════════════════
# Real-time Priorities: For time-critical applications
# ═══════════════════════════════════════════════════════════════
 
# Set real-time scheduling policy (requires CAP_SYS_NICE or root)
# SCHED_FIFO: First-in, first-out real-time scheduling
sudo chrt -f 50 ./realtime_task &
RT_PID=$!
echo "Started real-time task with priority 50 (PID: $RT_PID)"
 
# View scheduling policy and priority
chrt -p $RT_PID
 
# ═══════════════════════════════════════════════════════════════
# Control Groups (cgroups v2): Hierarchical resource control
# ═══════════════════════════════════════════════════════════════
 
# Create a CPU-limited cgroup (requires root, cgroups v2 mounted)
CGROUP_PATH="/sys/fs/cgroup/background_tasks"
sudo mkdir -p "$CGROUP_PATH"
 
# Set CPU weight (default 100, range 1-10000)
# Lower weight = less CPU share relative to other cgroups
echo 50 | sudo tee "$CGROUP_PATH/cpu.weight"
 
# Move a process into the cgroup
echo $NICE_PID | sudo tee "$CGROUP_PATH/cgroup.procs"
 
echo "Moved PID $NICE_PID to background_tasks cgroup with CPU weight 50"
 
# ═══════════════════════════════════════════════════════════════
# ionice: I/O Priority (separate from CPU priority)
# ═══════════════════════════════════════════════════════════════
 
# Classes: 1=realtime, 2=best-effort, 3=idle
# Within best-effort: 0 (highest) to 7 (lowest)
ionice -c 3 ./disk_heavy_job &
IO_PID=$!
echo "Started disk job with idle I/O priority (PID: $IO_PID)"
 
# View I/O priority
ionice -p $IO_PID
 
# ═══════════════════════════════════════════════════════════════
# Combined Priority Strategy: Real-world configuration
# ═══════════════════════════════════════════════════════════════
 
# For a background batch job:
# - High nice value (low CPU priority)
# - Idle I/O class (yields to interactive I/O)
# - Restricted to specific CPUs
nice -n 19 ionice -c 3 taskset -c 0-1 ./batch_job &
BATCH_PID=$!
echo "Started batch job with all priority restrictions (PID: $BATCH_PID)"
 
# Monitor the processes
echo ""
echo "Process priorities:"
ps -eo pid,ni,cls,pri,comm | grep -E "PID|cpu_intensive|realtime|disk_heavy|batch"

Priority Inheritance from Parent Processes

By default, child processes inherit their parent's base priority. This creates natural priority hierarchies—a low-priority daemon's children remain low priority. However, some system calls allow specifying child priority explicitly, and schedulers may reset priority for certain process types (e.g., setuid programs).

Priority Assignment in Linux

Linux employs a sophisticated multi-layered priority system that accommodates both traditional time-sharing workloads and real-time requirements. Understanding Linux priorities requires grasping the relationship between nice values, scheduling policies, and the underlying priority numbers.

The Linux Priority Hierarchy

Linux uses a 140-level priority scale internally:

Priorities 0-99: Reserved for real-time processes (SCHED_FIFO and SCHED_RR)
Priorities 100-139: For normal processes (SCHED_OTHER/SCHED_NORMAL)

The mapping is:

Real-time priority 99 → Internal priority 0 (highest)
Real-time priority 1  → Internal priority 98
Real-time priority 0  → Internal priority 99
Nice -20              → Internal priority 100
Nice 0                → Internal priority 120
Nice +19              → Internal priority 139 (lowest)

Converting Mermaid diagram...

Scheduling Policies and Their Priority Semantics

Linux supports multiple scheduling policies, each with different priority interpretations:

Linux Scheduling Policies
Policy	Type	Priority Range	Behavior	Use Case
SCHED_FIFO	Real-time	1-99	Runs until blocks or yields; no time slicing	Hard real-time, latency-critical
SCHED_RR	Real-time	1-99	Round-robin among same-priority RT processes	Real-time with fairness among equals
SCHED_DEADLINE	Real-time	Implicit (deadlines)	Earliest Deadline First (EDF)	Multimedia, periodic tasks
SCHED_OTHER	Normal	Nice -20 to +19	CFS with virtual runtime	General-purpose, default
SCHED_BATCH	Normal	Nice -20 to +19	CFS optimized for throughput	Batch processing, compilers
SCHED_IDLE	Normal	N/A (lowest)	Only runs when CPU idle	Background maintenance

linux_priority_inspection.c
C
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
#define _GNU_SOURCE
#include <sched.h>
#include <sys/resource.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <string.h>
#include <errno.h>
 
/**
 * Demonstrates Linux priority inspection and modification.
 * Compile: gcc -o priority_demo linux_priority_inspection.c
 * Run: ./priority_demo (some operations require root)
 */
 
const char* policy_name(int policy) {
    switch (policy) {
        case SCHED_OTHER:   return "SCHED_OTHER (Normal/CFS)";
        case SCHED_FIFO:    return "SCHED_FIFO (Real-time FIFO)";
        case SCHED_RR:      return "SCHED_RR (Real-time Round-Robin)";
        case SCHED_BATCH:   return "SCHED_BATCH (Batch processing)";
        case SCHED_IDLE:    return "SCHED_IDLE (Idle priority)";
#ifdef SCHED_DEADLINE
        case SCHED_DEADLINE: return "SCHED_DEADLINE (EDF)";
#endif
        default:            return "Unknown";
    }
}
 
void print_priority_info(pid_t pid) {
    struct sched_param param;
    int policy;
    int nice_val;
    
    printf("\n═══════════════════════════════════════════════════════════════\n");
    printf("Priority Information for PID %d\n", pid);
    printf("═══════════════════════════════════════════════════════════════\n");
    
    // Get scheduling policy
    policy = sched_getscheduler(pid);
    if (policy == -1) {
        perror("sched_getscheduler failed");
        return;
    }
    printf("Scheduling Policy: %s\n", policy_name(policy));
    
    // Get scheduling parameters (priority for RT policies)
    if (sched_getparam(pid, &param) == -1) {
        perror("sched_getparam failed");
        return;
    }
    printf("Scheduler Priority: %d\n", param.sched_priority);
    
    // Get nice value (relevant for SCHED_OTHER, SCHED_BATCH)
    errno = 0;
    nice_val = getpriority(PRIO_PROCESS, pid);
    if (errno != 0) {
        perror("getpriority failed");
    } else {
        printf("Nice Value: %d\n", nice_val);
    }
    
    // Print valid priority ranges
    printf("\nValid priority ranges:\n");
    printf("  SCHED_FIFO/%s: %d to %d\n", 
           "SCHED_RR",
           sched_get_priority_min(SCHED_FIFO),
           sched_get_priority_max(SCHED_FIFO));
    printf("  SCHED_OTHER: Nice values -20 to +19\n");
}
 
int main(int argc, char *argv[]) {
    pid_t target_pid;
    
    if (argc > 1) {
        target_pid = atoi(argv[1]);
    } else {
        target_pid = getpid();  // Self
    }
    
    // Print current priority info
    print_priority_info(target_pid);
    
    // Demonstrate nice value modification (if examining self)
    if (target_pid == getpid()) {
        printf("\n─ Demonstrating priority modifications ─\n");
        
        // Increase nice value (lower priority) - always succeeds
        if (nice(5) == -1 && errno != 0) {
            perror("nice(5) failed");
        } else {
            printf("After nice(5): Nice value = %d\n", 
                   getpriority(PRIO_PROCESS, 0));
        }
        
        // Attempt to decrease nice value (raise priority)
        // Requires CAP_SYS_NICE or root
        errno = 0;
        if (nice(-10) == -1 && errno != 0) {
            printf("nice(-10) failed: %s\n", strerror(errno));
            printf("(Requires root or CAP_SYS_NICE to decrease nice value)\n");
        } else {
            printf("After nice(-10): Nice value = %d\n",
                   getpriority(PRIO_PROCESS, 0));
        }
        
        // Attempt to switch to real-time scheduling
        struct sched_param rt_param;
        rt_param.sched_priority = 50;
        
        if (sched_setscheduler(0, SCHED_FIFO, &rt_param) == -1) {
            printf("sched_setscheduler(SCHED_FIFO, 50) failed: %s\n", 
                   strerror(errno));
            printf("(Requires root or CAP_SYS_NICE for real-time scheduling)\n");
        } else {
            printf("Successfully switched to SCHED_FIFO with priority 50\n");
            print_priority_info(getpid());
        }
    }
    
    return 0;
}

Real-Time Priority Dangers

A runaway SCHED_FIFO process at high priority can render a system unresponsive since it will never yield to lower-priority processes (including the shell). Always have a rescue mechanism: a separate real-time watchdog, SSH access from another machine, or the kernel's RT throttling feature (sched_rt_runtime_us and sched_rt_period_us).

Priority Assignment in Windows

Windows implements a two-level priority system that separates process-level priority classes from thread-level relative priorities. The combination determines the thread's base priority, which the scheduler then dynamically adjusts.

Priority Classes and Thread Priorities

Windows uses a 0-31 priority range where higher numbers indicate higher priority:

Priorities 16-31: Real-time priority range
Priorities 1-15: Dynamic priority range (normal applications)
Priority 0: Reserved for the zero-page thread

Windows Base Priority Calculation Matrix
Process Class →	Idle	Below Normal	Normal	Above Normal	High	Realtime
Thread: Time Critical	15	15	15	15	15	31
Thread: Highest	6	8	10	12	15	26
Thread: Above Normal	5	7	9	11	14	25
Thread: Normal	4	6	8	10	13	24
Thread: Below Normal	3	5	7	9	12	23
Thread: Lowest	2	4	6	8	11	22
Thread: Idle	1	1	1	1	1	16

Base Priority vs Current Priority

Windows distinguishes between:

Base Priority: Computed from process class and thread priority, remains stable
Current Priority: Dynamically boosted above base priority based on events

The current priority can temporarily exceed the base priority but never falls below it. After a boost, priority decays by 1 each quantum until returning to base.

Priority Boost Triggers

Windows boosts thread priorities in several scenarios:

Windows Priority Boost Events

•I/O Completion: Thread wakes from I/O wait. Boost magnitude depends on device type (keyboard: larger boost; disk: smaller boost).
•Semaphore/Event Release: Thread unblocked by synchronization primitive receives a boost.
•Foreground Window: Threads of the foreground application receive increased quantum and potential priority boost.
•GUI Input: Threads processing mouse or keyboard input receive temporary boosts.
•Priority Inversion Mitigation: Windows may boost waiting thread priority when it detects a thread holding a needed resource.
•MMCSS (Multimedia Class Scheduler Service): Audio/video threads registered with MMCSS receive guaranteed CPU time and elevated priority.

windows_priority.cpp
C++ (Windows)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
#include <windows.h>
#include <stdio.h>
#include <avrt.h>  // For MMCSS
 
#pragma comment(lib, "avrt.lib")
 
/**
 * Demonstrates Windows priority mechanisms including:
 * - Process priority classes
 * - Thread priorities
 * - Priority boosting control
 * - Multimedia Class Scheduler Service (MMCSS)
 */
 
void PrintPriorityInfo(HANDLE hThread, const char* label) {
    printf("\n%s:\n", label);
    printf("  Thread Priority: %d\n", GetThreadPriority(hThread));
    
    // Note: GetThreadPriorityBoost returns whether boosting is DISABLED
    BOOL boostDisabled;
    if (GetThreadPriorityBoost(hThread, &boostDisabled)) {
        printf("  Priority Boost: %s\n", 
               boostDisabled ? "Disabled" : "Enabled");
    }
}
 
const char* PriorityClassName(DWORD priorityClass) {
    switch (priorityClass) {
        case IDLE_PRIORITY_CLASS:         return "IDLE";
        case BELOW_NORMAL_PRIORITY_CLASS: return "BELOW_NORMAL";
        case NORMAL_PRIORITY_CLASS:       return "NORMAL";
        case ABOVE_NORMAL_PRIORITY_CLASS: return "ABOVE_NORMAL";
        case HIGH_PRIORITY_CLASS:         return "HIGH";
        case REALTIME_PRIORITY_CLASS:     return "REALTIME";
        default:                          return "UNKNOWN";
    }
}
 
int main() {
    HANDLE hProcess = GetCurrentProcess();
    HANDLE hThread = GetCurrentThread();
    
    printf("═══════════════════════════════════════════════════════════════\n");
    printf("Windows Priority Demonstration\n");
    printf("═══════════════════════════════════════════════════════════════\n");
    
    // ─────────────────────────────────────────────────────────────────
    // Display current priority settings
    // ─────────────────────────────────────────────────────────────────
    DWORD priorityClass = GetPriorityClass(hProcess);
    printf("\nInitial Process Priority Class: %s\n", 
           PriorityClassName(priorityClass));
    PrintPriorityInfo(hThread, "Initial Thread State");
    
    // ─────────────────────────────────────────────────────────────────
    // Modify process priority class
    // ─────────────────────────────────────────────────────────────────
    printf("\n─ Setting process to ABOVE_NORMAL_PRIORITY_CLASS ─\n");
    if (!SetPriorityClass(hProcess, ABOVE_NORMAL_PRIORITY_CLASS)) {
        printf("SetPriorityClass failed: %lu\n", GetLastError());
    } else {
        printf("Process Priority Class: %s\n",
               PriorityClassName(GetPriorityClass(hProcess)));
        PrintPriorityInfo(hThread, "After Process Priority Change");
    }
    
    // ─────────────────────────────────────────────────────────────────
    // Modify thread priority within class
    // ─────────────────────────────────────────────────────────────────
    printf("\n─ Setting thread to THREAD_PRIORITY_HIGHEST ─\n");
    if (!SetThreadPriority(hThread, THREAD_PRIORITY_HIGHEST)) {
        printf("SetThreadPriority failed: %lu\n", GetLastError());
    } else {
        PrintPriorityInfo(hThread, "After Thread Priority Change");
    }
    
    // ─────────────────────────────────────────────────────────────────
    // Disable priority boosting
    // ─────────────────────────────────────────────────────────────────
    printf("\n─ Disabling priority boosting ─\n");
    if (!SetThreadPriorityBoost(hThread, TRUE)) {  // TRUE = disable boost
        printf("SetThreadPriorityBoost failed: %lu\n", GetLastError());
    } else {
        PrintPriorityInfo(hThread, "With Boosting Disabled");
    }
    
    // ─────────────────────────────────────────────────────────────────
    // Demonstrate MMCSS for multimedia applications
    // ─────────────────────────────────────────────────────────────────
    printf("\n─ Registering with Multimedia Class Scheduler Service ─\n");
    DWORD taskIndex = 0;
    HANDLE hTask = AvSetMmThreadCharacteristics(TEXT("Pro Audio"), &taskIndex);
    
    if (hTask == NULL) {
        printf("AvSetMmThreadCharacteristics failed: %lu\n", GetLastError());
        printf("(This may require the MMCSS service to be running)\n");
    } else {
        printf("Successfully registered as 'Pro Audio' thread\n");
        printf("Task Index: %lu\n", taskIndex);
        PrintPriorityInfo(hThread, "With MMCSS Registration");
        
        // MMCSS-registered threads can request priority changes
        if (AvSetMmThreadPriority(hTask, AVRT_PRIORITY_CRITICAL)) {
            printf("Set MMCSS priority to CRITICAL\n");
            PrintPriorityInfo(hThread, "With MMCSS Critical Priority");
        }
        
        // Unregister when done
        AvRevertMmThreadCharacteristics(hTask);
        printf("\nMCSS registration reverted\n");
    }
    
    // ─────────────────────────────────────────────────────────────────
    // Demonstrate real-time priority (requires elevation)
    // ─────────────────────────────────────────────────────────────────
    printf("\n─ Attempting REALTIME_PRIORITY_CLASS ─\n");
    if (!SetPriorityClass(hProcess, REALTIME_PRIORITY_CLASS)) {
        DWORD error = GetLastError();
        printf("SetPriorityClass(REALTIME) failed: %lu\n", error);
        if (error == ERROR_ACCESS_DENIED) {
            printf("(Requires administrator privileges)\n");
        }
    } else {
        printf("Process Priority Class: %s\n",
               PriorityClassName(GetPriorityClass(hProcess)));
        printf("WARNING: REALTIME_PRIORITY_CLASS can destabilize the system!\n");
        
        // Immediately step back from realtime
        SetPriorityClass(hProcess, NORMAL_PRIORITY_CLASS);
        printf("Reverted to NORMAL_PRIORITY_CLASS\n");
    }
    
    return 0;
}

MMCSS for Audio/Video Applications

For multimedia applications requiring guaranteed CPU time, the Multimedia Class Scheduler Service (MMCSS) provides a Windows-native solution. By registering as 'Pro Audio' or 'Playback', threads receive priority boosts and protected time slices. This is preferred over manually setting REALTIME_PRIORITY_CLASS, which can destabilize system-critical services.

Priority Assignment Best Practices

Incorrect priority assignment leads to performance degradation, starvation, priority inversion, and system instability. The following best practices distill decades of operating systems experience into actionable guidelines:

Priority Assignment Guidelines

•Default to Default Priority — Unless you have a specific, measurable reason to change priorities, use the system defaults. Priority elevation should be a last resort, not a first instinct.
•Raise Priority Sparingly, Lower Liberally — It's almost always safe to lower your process's priority (increase nice value). Raising priority risks starving other processes and should require explicit justification.
•Prefer Time Limits to Priority Boosts — Instead of raising priority to complete work faster, consider whether the work can be structured with reasonable time budgets at default priority.
•Match Priority to Temporal Requirements — Use high priority only for latency-sensitive work with genuine real-time constraints (audio playback, control systems). Throughput-oriented work (compilation, data processing) rarely needs elevated priority.
•Avoid Priority Starvation by Design — If using static priorities, ensure lower-priority processes have guaranteed execution windows. Use aging, deadline scheduling, or priority ceiling protocols.
•Consider Priority Inheritance in Lock Design — When high-priority threads may wait for locks held by low-priority threads, implement priority inheritance to avoid unbounded priority inversion.
•Test Under Contention — Priority-related bugs often only manifest under heavy system load. Test with realistic workloads that create scheduler pressure.
•Document Priority Decisions — Priority assignments are often made once and forgotten. Document why specific priorities were chosen so future maintainers understand the constraints.

• Lower priority for batch jobs • Use system-provided mechanisms (MMCSS, cgroups) • Test priority changes under load • Implement priority inheritance for locks • Use deadlines instead of static priorities when possible

Don't

• Set real-time priority without thorough analysis • Assume high priority means faster completion • Ignore priority on multi-tenant systems • Create priority inversion scenarios • Use priority elevation as first solution to performance issues

Summary: Priority Assignment Fundamentals

We have explored the foundational concepts of priority assignment in operating systems. The key takeaways form the basis for understanding the more advanced topics in the remainder of this module:

Key Takeaways

•Priority semantics vary by OS — Different operating systems use opposite numerical conventions. Always verify whether higher numbers mean higher or lower priority.
•Static priorities offer predictability — Fixed priorities, once assigned, never change. This is essential for real-time systems but risks starvation.
•Dynamic priorities offer adaptability — Priorities that adjust based on runtime behavior improve fairness and responsiveness but sacrifice determinism.
•Priority sources are internal and external — The OS derives priorities from observed behavior (internal) and from user/application specifications (external).
•Linux uses a 140-level hybrid system — Combining real-time scheduling policies (0-99) with normal scheduling (nice values mapping to 100-139).
•Windows uses priority class + thread priority — The combination produces a base priority, which the system dynamically boosts based on events.
•Priority assignment requires discipline — Elevating priority should be a last resort with documented justification. Default priorities are usually correct.

What's Next:

Now that we understand how priorities are assigned, we'll examine how the scheduler uses these priorities to make scheduling decisions. The next page explores preemptive versus non-preemptive priority scheduling—understanding when a higher-priority process can interrupt a running lower-priority process, and the implications of each approach for system behavior.

Page Complete

You now understand the fundamental mechanisms of priority assignment in operating systems, including static vs dynamic priorities, internal vs external sources, and the specific implementations in Linux and Windows. You are prepared to analyze how these priorities influence scheduling decisions.

1 / 5

Loading learning content...

Operating SystemsCPU Scheduling

Priority Scheduling

LevelIntermediate

Duration90 mins

TopicCPU Scheduling

1 / 5

Priority Assignment

The Priority Imperative in Modern Computing

What You Will Learn

Foundations of Priority in Operating Systems

The Semantic Challenge of Priority

Before examining mechanisms, we must address a fundamental semantic ambiguity: what does a priority number actually mean? Different operating systems adopt diametrically opposite conventions:

Lower number = Higher priority: Used in UNIX-like systems (Linux, macOS, BSD). A process with priority 0 is more important than one with priority 10.
Higher number = Higher priority: Used in Windows and many real-time operating systems. A thread with priority 31 runs before one with priority 15.

This inconsistency has caused countless bugs when developers move between platforms. The conceptual priority (importance) maps to numerical priority differently depending on the operating system.

Priority Convention Comparison Across Operating Systems
Operating System	Priority Range	Convention	Highest Priority Value	Lowest Priority Value
Linux (Nice values)	-20 to +19	Lower = Higher	-20	+19
Linux (Real-time)	0 to 99	Higher = Higher	99	0
Windows	0 to 31	Higher = Higher	31 (Real-time)	0 (Idle)
macOS (Mach)	0 to 127	Lower = Higher	0	127
FreeBSD	-20 to +20	Lower = Higher	-20	+20
VxWorks (RTOS)	0 to 255	Lower = Higher	0	255
FreeRTOS	0 to configMAX_PRIORITIES-1	Higher = Higher	configMAX_PRIORITIES-1	0

Priority Inversion of Semantics

Priority as a Scheduling Parameter

Mathematically, priority functions as a partial ordering over the set of ready processes. Given processes P₁, P₂, ..., Pₙ in the ready queue, the scheduler selects process Pᵢ such that:

priority(Pᵢ) ≥ priority(Pⱼ) for all j ≠ i

Or, in systems where lower values indicate higher priority:

priority(Pᵢ) ≤ priority(Pⱼ) for all j ≠ i

This ordering determines CPU allocation decisions, but the derivation of priority values—how they are assigned and updated—varies dramatically across scheduling implementations.

Static Priority Assignment

Characteristics of Static Priority

Static priority systems exhibit several defining characteristics:

Determinism: Given initial priority assignments, the scheduling sequence is fully predictable. This property is crucial for real-time systems where timing guarantees must be made.
Simplicity: No runtime computation is required to update priorities. The scheduler simply compares fixed values.
Stability: Process behavior cannot influence its scheduling priority. A process cannot 'game' the system by altering its execution patterns.
Potential for Starvation: Low-priority processes may never execute if higher-priority processes remain runnable indefinitely.

Common Static Priority Assignment Strategies

•Process Type Classification — System processes receive higher priority than user processes. Kernel threads outrank application threads. Device drivers may occupy intermediate levels.
•User Specification — Unix nice values allow users to voluntarily lower their process priority (or raise it, if privileged). Windows allows priority class specification via SetPriorityClass().
•Application-Defined Priority — Software architects designate priority levels during design. A web server's request handler might run at higher priority than its logging subsystem.
•Resource Ownership — Processes holding critical resources (locks, device handles) may receive elevated priority to release those resources quickly.
•Rate Monotonic Assignment — In real-time systems, processes with shorter periods receive higher priority. This is provably optimal for certain task sets.

static_priority_examples.c
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
#include <sched.h>
#include <pthread.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <errno.h>
 
/**
 * Demonstrates static priority assignment using POSIX real-time scheduling.
 * Real-time priorities in Linux range from 1 to 99, with 99 being highest.
 * Requires CAP_SYS_NICE capability or root privileges.
 */
 
void* high_priority_task(void* arg) {
    printf("High priority task executing (priority 80)\n");
    // Critical real-time work here
    volatile long sum = 0;
    for (long i = 0; i < 100000000; i++) {
        sum += i;
    }
    printf("High priority task completed\n");
    return NULL;
}
 
void* low_priority_task(void* arg) {
    printf("Low priority task executing (priority 20)\n");
    // Less critical work here
    volatile long sum = 0;
    for (long i = 0; i < 100000000; i++) {
        sum += i;
    }
    printf("Low priority task completed\n");
    return NULL;
}
 
int main() {
    pthread_t high_thread, low_thread;
    pthread_attr_t high_attr, low_attr;
    struct sched_param high_param, low_param;
    
    // Initialize thread attributes
    pthread_attr_init(&high_attr);
    pthread_attr_init(&low_attr);
    
    // Set scheduling policy to SCHED_FIFO (real-time, static priority)
    pthread_attr_setschedpolicy(&high_attr, SCHED_FIFO);
    pthread_attr_setschedpolicy(&low_attr, SCHED_FIFO);
    
    // Enable explicit scheduling (don't inherit from parent)
    pthread_attr_setinheritsched(&high_attr, PTHREAD_EXPLICIT_SCHED);
    pthread_attr_setinheritsched(&low_attr, PTHREAD_EXPLICIT_SCHED);
    
    // Assign static priorities
    high_param.sched_priority = 80;  // High priority (1-99 range)
    low_param.sched_priority = 20;   // Low priority
    
    pthread_attr_setschedparam(&high_attr, &high_param);
    pthread_attr_setschedparam(&low_attr, &low_param);
    
    // Create threads with static priorities
    // Note: Low priority thread created first, but high priority runs first
    if (pthread_create(&low_thread, &low_attr, low_priority_task, NULL) != 0) {
        perror("Failed to create low priority thread");
        fprintf(stderr, "Error: %s\n", strerror(errno));
        fprintf(stderr, "Hint: Run with sudo or set CAP_SYS_NICE\n");
        exit(1);
    }
    
    if (pthread_create(&high_thread, &high_attr, high_priority_task, NULL) != 0) {
        perror("Failed to create high priority thread");
        exit(1);
    }
    
    // Wait for completion
    pthread_join(high_thread, NULL);
    pthread_join(low_thread, NULL);
    
    // Clean up
    pthread_attr_destroy(&high_attr);
    pthread_attr_destroy(&low_attr);
    
    printf("Both tasks completed\n");
    return 0;
}

Rate Monotonic Scheduling (RMS)

Dynamic Priority Assignment

The Need for Dynamic Priorities

Static priorities, while simple, cannot account for:

Behavioral Changes: A process may alternate between CPU-intensive and I/O-intensive phases. Optimal priority differs between phases.
System Load Variations: The 'right' priority for a process depends on what else is running. A background task might safely run at higher priority when the system is idle.
Fairness Requirements: Without dynamic adjustment, high-priority processes can monopolize the CPU, starving lower-priority work.
Responsiveness Demands: Interactive processes need priority boosts when users interact with them, even if they previously ran at lower priority.

Dynamic Priority Computation

Dynamic priorities are typically computed as a function of multiple factors:

dynamic_priority_formula.txt

Mathematical Model

Dynamic Priority Computation (General Model)
═══════════════════════════════════════════════════════════════════════
 
    effective_priority(P, t) = base_priority(P) 
                              + f(cpu_usage(P, t)) 
                              + g(wait_time(P, t))
                              + h(io_behavior(P, t))
                              + bonus(P, t)
 
Where:
────────────────────────────────────────────────────────────────────────
• base_priority(P)     : Initial priority (may be user-specified)
• cpu_usage(P, t)      : Recent CPU consumption (typically decreases priority)
• wait_time(P, t)      : Time spent waiting (may increase priority for fairness)
• io_behavior(P, t)    : I/O-bound processes often receive priority boosts
• bonus(P, t)          : Context-dependent adjustments (interactive use, etc.)
 
Linux CFS Example (Simplified Virtual Runtime):
────────────────────────────────────────────────────────────────────────
    vruntime(P) = vruntime(P) + (delta_exec * NICE_0_LOAD) / weight(P)
    
    Where weight(P) is derived from nice value:
        weight(nice) = 1024 / (1.25 ^ nice)
    
    Process with LOWEST vruntime runs next (most "starved" of CPU time).
 
Windows Priority Boost Example:
────────────────────────────────────────────────────────────────────────
    When I/O completes:
        boosted_priority = base_priority + io_boost
        (io_boost depends on device type: keyboard > disk > network)
    
    Boost decays: After each quantum without blocking:
        priority = max(priority - 1, base_priority)
 
FreeBSD ULE Scheduler:
────────────────────────────────────────────────────────────────────────
    priority = PRI_MIN_TIMESHARE 
               + (recent_cpu_ticks * INVERSE_ESTCPU_WEIGHT)
               - (interactivity_score * INVERSE_INTERACT_WEIGHT)
    
    Interactivity score rewards processes that sleep frequently.

Factors Influencing Dynamic Priority

•CPU Consumption History — Processes that have consumed significant CPU time recently see priority reductions. This prevents CPU-bound processes from monopolizing the processor.
•Time Spent Waiting — Processes waiting in the ready queue may receive priority increases (aging). This prevents indefinite starvation of low-priority processes.
•I/O Behavior Patterns — I/O-bound processes typically receive priority boosts. They voluntarily yield the CPU while waiting for I/O, so boosting their priority when ready improves overall system throughput.
•Interactive Events — Keyboard input, mouse clicks, and window focus changes may trigger priority boosts for the associated process, improving perceived responsiveness.
•Real-time Constraints — Approaching deadlines may increase priority in deadline-aware schedulers. Missing a deadline may decrease priority or trigger rescheduling.
•Resource Contention — Processes holding locks or other synchronization primitives may receive temporary boosts to release those resources quickly, reducing overall blocking time.

Decay Functions and Temporal Relevance

Internal vs External Priority Sources

Internal Priority Factors

Internal factors are computed automatically by the operating system:

Internal Priority Factors
Factor	Measurement	Typical Effect	Rationale
CPU Burst Length	Runtime between I/O operations	Shorter bursts → higher priority	Favor I/O-bound processes for throughput
I/O Wait Ratio	Time waiting for I/O / total time	Higher ratio → higher priority	I/O-bound processes release CPU voluntarily
Memory Usage	Working set size, page fault rate	Varies by policy	May penalize memory-hungry processes
Time in System	Total elapsed time since creation	Longer → higher priority (aging)	Prevent indefinite starvation
Sleep Frequency	How often process voluntarily yields	More sleep → higher priority	Reward cooperative behavior
Recent CPU Time	CPU seconds in last N milliseconds	More CPU → lower priority	Share CPU fairly among processes

External Priority Factors

External factors come from outside the kernel's observations:

External Priority Factors
Factor	Source	Mechanism	Example
User-specified Nice Value	User or script	`nice`, `renice` commands	`nice -n 10 ./background_job`
Process Priority Class	Application API	System call at creation	`SetPriorityClass()` on Windows
Real-time Classification	Administrator/application	Scheduler policy selection	`sched_setscheduler()` with `SCHED_FIFO`
Control Group (cgroup) Weights	System administrator	cgroup configuration files	Set cpu.weight in cgroup v2
Container/VM Resource Limits	Orchestration system	Container runtime configuration	Kubernetes resource requests/limits
Affinity Settings	Application or admin	Processor affinity masks	`taskset` command or `pthread_setaffinity_np()`

priority_sources.sh
Shell
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
#!/bin/bash
# Demonstrating various external priority mechanisms on Linux
 
# ═══════════════════════════════════════════════════════════════
# Nice Values: Traditional UNIX priority adjustment
# ═══════════════════════════════════════════════════════════════
 
# Start a process with reduced priority (nice value +10)
nice -n 10 ./cpu_intensive_job &
NICE_PID=$!
echo "Started CPU job with nice 10 (PID: $NICE_PID)"
 
# Adjust priority of running process
renice +15 -p $NICE_PID
echo "Adjusted nice value to 15"
 
# View nice values of all user processes
ps -eo pid,ni,comm --sort=-ni | head -20
 
# ═══════════════════════════════════════════════════════════════
# Real-time Priorities: For time-critical applications
# ═══════════════════════════════════════════════════════════════
 
# Set real-time scheduling policy (requires CAP_SYS_NICE or root)
# SCHED_FIFO: First-in, first-out real-time scheduling
sudo chrt -f 50 ./realtime_task &
RT_PID=$!
echo "Started real-time task with priority 50 (PID: $RT_PID)"
 
# View scheduling policy and priority
chrt -p $RT_PID
 
# ═══════════════════════════════════════════════════════════════
# Control Groups (cgroups v2): Hierarchical resource control
# ═══════════════════════════════════════════════════════════════
 
# Create a CPU-limited cgroup (requires root, cgroups v2 mounted)
CGROUP_PATH="/sys/fs/cgroup/background_tasks"
sudo mkdir -p "$CGROUP_PATH"
 
# Set CPU weight (default 100, range 1-10000)
# Lower weight = less CPU share relative to other cgroups
echo 50 | sudo tee "$CGROUP_PATH/cpu.weight"
 
# Move a process into the cgroup
echo $NICE_PID | sudo tee "$CGROUP_PATH/cgroup.procs"
 
echo "Moved PID $NICE_PID to background_tasks cgroup with CPU weight 50"
 
# ═══════════════════════════════════════════════════════════════
# ionice: I/O Priority (separate from CPU priority)
# ═══════════════════════════════════════════════════════════════
 
# Classes: 1=realtime, 2=best-effort, 3=idle
# Within best-effort: 0 (highest) to 7 (lowest)
ionice -c 3 ./disk_heavy_job &
IO_PID=$!
echo "Started disk job with idle I/O priority (PID: $IO_PID)"
 
# View I/O priority
ionice -p $IO_PID
 
# ═══════════════════════════════════════════════════════════════
# Combined Priority Strategy: Real-world configuration
# ═══════════════════════════════════════════════════════════════
 
# For a background batch job:
# - High nice value (low CPU priority)
# - Idle I/O class (yields to interactive I/O)
# - Restricted to specific CPUs
nice -n 19 ionice -c 3 taskset -c 0-1 ./batch_job &
BATCH_PID=$!
echo "Started batch job with all priority restrictions (PID: $BATCH_PID)"
 
# Monitor the processes
echo ""
echo "Process priorities:"
ps -eo pid,ni,cls,pri,comm | grep -E "PID|cpu_intensive|realtime|disk_heavy|batch"

Priority Inheritance from Parent Processes

Priority Assignment in Linux

The Linux Priority Hierarchy

Linux uses a 140-level priority scale internally:

Priorities 0-99: Reserved for real-time processes (SCHED_FIFO and SCHED_RR)
Priorities 100-139: For normal processes (SCHED_OTHER/SCHED_NORMAL)

The mapping is:

Real-time priority 99 → Internal priority 0 (highest)
Real-time priority 1  → Internal priority 98
Real-time priority 0  → Internal priority 99
Nice -20              → Internal priority 100
Nice 0                → Internal priority 120
Nice +19              → Internal priority 139 (lowest)

Converting Mermaid diagram...

Scheduling Policies and Their Priority Semantics

Linux supports multiple scheduling policies, each with different priority interpretations:

Linux Scheduling Policies
Policy	Type	Priority Range	Behavior	Use Case
SCHED_FIFO	Real-time	1-99	Runs until blocks or yields; no time slicing	Hard real-time, latency-critical
SCHED_RR	Real-time	1-99	Round-robin among same-priority RT processes	Real-time with fairness among equals
SCHED_DEADLINE	Real-time	Implicit (deadlines)	Earliest Deadline First (EDF)	Multimedia, periodic tasks
SCHED_OTHER	Normal	Nice -20 to +19	CFS with virtual runtime	General-purpose, default
SCHED_BATCH	Normal	Nice -20 to +19	CFS optimized for throughput	Batch processing, compilers
SCHED_IDLE	Normal	N/A (lowest)	Only runs when CPU idle	Background maintenance

linux_priority_inspection.c
C
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
#define _GNU_SOURCE
#include <sched.h>
#include <sys/resource.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <string.h>
#include <errno.h>
 
/**
 * Demonstrates Linux priority inspection and modification.
 * Compile: gcc -o priority_demo linux_priority_inspection.c
 * Run: ./priority_demo (some operations require root)
 */
 
const char* policy_name(int policy) {
    switch (policy) {
        case SCHED_OTHER:   return "SCHED_OTHER (Normal/CFS)";
        case SCHED_FIFO:    return "SCHED_FIFO (Real-time FIFO)";
        case SCHED_RR:      return "SCHED_RR (Real-time Round-Robin)";
        case SCHED_BATCH:   return "SCHED_BATCH (Batch processing)";
        case SCHED_IDLE:    return "SCHED_IDLE (Idle priority)";
#ifdef SCHED_DEADLINE
        case SCHED_DEADLINE: return "SCHED_DEADLINE (EDF)";
#endif
        default:            return "Unknown";
    }
}
 
void print_priority_info(pid_t pid) {
    struct sched_param param;
    int policy;
    int nice_val;
    
    printf("\n═══════════════════════════════════════════════════════════════\n");
    printf("Priority Information for PID %d\n", pid);
    printf("═══════════════════════════════════════════════════════════════\n");
    
    // Get scheduling policy
    policy = sched_getscheduler(pid);
    if (policy == -1) {
        perror("sched_getscheduler failed");
        return;
    }
    printf("Scheduling Policy: %s\n", policy_name(policy));
    
    // Get scheduling parameters (priority for RT policies)
    if (sched_getparam(pid, &param) == -1) {
        perror("sched_getparam failed");
        return;
    }
    printf("Scheduler Priority: %d\n", param.sched_priority);
    
    // Get nice value (relevant for SCHED_OTHER, SCHED_BATCH)
    errno = 0;
    nice_val = getpriority(PRIO_PROCESS, pid);
    if (errno != 0) {
        perror("getpriority failed");
    } else {
        printf("Nice Value: %d\n", nice_val);
    }
    
    // Print valid priority ranges
    printf("\nValid priority ranges:\n");
    printf("  SCHED_FIFO/%s: %d to %d\n", 
           "SCHED_RR",
           sched_get_priority_min(SCHED_FIFO),
           sched_get_priority_max(SCHED_FIFO));
    printf("  SCHED_OTHER: Nice values -20 to +19\n");
}
 
int main(int argc, char *argv[]) {
    pid_t target_pid;
    
    if (argc > 1) {
        target_pid = atoi(argv[1]);
    } else {
        target_pid = getpid();  // Self
    }
    
    // Print current priority info
    print_priority_info(target_pid);
    
    // Demonstrate nice value modification (if examining self)
    if (target_pid == getpid()) {
        printf("\n─ Demonstrating priority modifications ─\n");
        
        // Increase nice value (lower priority) - always succeeds
        if (nice(5) == -1 && errno != 0) {
            perror("nice(5) failed");
        } else {
            printf("After nice(5): Nice value = %d\n", 
                   getpriority(PRIO_PROCESS, 0));
        }
        
        // Attempt to decrease nice value (raise priority)
        // Requires CAP_SYS_NICE or root
        errno = 0;
        if (nice(-10) == -1 && errno != 0) {
            printf("nice(-10) failed: %s\n", strerror(errno));
            printf("(Requires root or CAP_SYS_NICE to decrease nice value)\n");
        } else {
            printf("After nice(-10): Nice value = %d\n",
                   getpriority(PRIO_PROCESS, 0));
        }
        
        // Attempt to switch to real-time scheduling
        struct sched_param rt_param;
        rt_param.sched_priority = 50;
        
        if (sched_setscheduler(0, SCHED_FIFO, &rt_param) == -1) {
            printf("sched_setscheduler(SCHED_FIFO, 50) failed: %s\n", 
                   strerror(errno));
            printf("(Requires root or CAP_SYS_NICE for real-time scheduling)\n");
        } else {
            printf("Successfully switched to SCHED_FIFO with priority 50\n");
            print_priority_info(getpid());
        }
    }
    
    return 0;
}

Real-Time Priority Dangers

Priority Assignment in Windows

Priority Classes and Thread Priorities

Windows uses a 0-31 priority range where higher numbers indicate higher priority:

Priorities 16-31: Real-time priority range
Priorities 1-15: Dynamic priority range (normal applications)
Priority 0: Reserved for the zero-page thread

Windows Base Priority Calculation Matrix
Process Class →	Idle	Below Normal	Normal	Above Normal	High	Realtime
Thread: Time Critical	15	15	15	15	15	31
Thread: Highest	6	8	10	12	15	26
Thread: Above Normal	5	7	9	11	14	25
Thread: Normal	4	6	8	10	13	24
Thread: Below Normal	3	5	7	9	12	23
Thread: Lowest	2	4	6	8	11	22
Thread: Idle	1	1	1	1	1	16

Base Priority vs Current Priority

Windows distinguishes between:

Base Priority: Computed from process class and thread priority, remains stable
Current Priority: Dynamically boosted above base priority based on events

The current priority can temporarily exceed the base priority but never falls below it. After a boost, priority decays by 1 each quantum until returning to base.

Priority Boost Triggers

Windows boosts thread priorities in several scenarios:

Windows Priority Boost Events

•I/O Completion: Thread wakes from I/O wait. Boost magnitude depends on device type (keyboard: larger boost; disk: smaller boost).
•Semaphore/Event Release: Thread unblocked by synchronization primitive receives a boost.
•Foreground Window: Threads of the foreground application receive increased quantum and potential priority boost.
•GUI Input: Threads processing mouse or keyboard input receive temporary boosts.
•Priority Inversion Mitigation: Windows may boost waiting thread priority when it detects a thread holding a needed resource.
•MMCSS (Multimedia Class Scheduler Service): Audio/video threads registered with MMCSS receive guaranteed CPU time and elevated priority.

windows_priority.cpp
C++ (Windows)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
#include <windows.h>
#include <stdio.h>
#include <avrt.h>  // For MMCSS
 
#pragma comment(lib, "avrt.lib")
 
/**
 * Demonstrates Windows priority mechanisms including:
 * - Process priority classes
 * - Thread priorities
 * - Priority boosting control
 * - Multimedia Class Scheduler Service (MMCSS)
 */
 
void PrintPriorityInfo(HANDLE hThread, const char* label) {
    printf("\n%s:\n", label);
    printf("  Thread Priority: %d\n", GetThreadPriority(hThread));
    
    // Note: GetThreadPriorityBoost returns whether boosting is DISABLED
    BOOL boostDisabled;
    if (GetThreadPriorityBoost(hThread, &boostDisabled)) {
        printf("  Priority Boost: %s\n", 
               boostDisabled ? "Disabled" : "Enabled");
    }
}
 
const char* PriorityClassName(DWORD priorityClass) {
    switch (priorityClass) {
        case IDLE_PRIORITY_CLASS:         return "IDLE";
        case BELOW_NORMAL_PRIORITY_CLASS: return "BELOW_NORMAL";
        case NORMAL_PRIORITY_CLASS:       return "NORMAL";
        case ABOVE_NORMAL_PRIORITY_CLASS: return "ABOVE_NORMAL";
        case HIGH_PRIORITY_CLASS:         return "HIGH";
        case REALTIME_PRIORITY_CLASS:     return "REALTIME";
        default:                          return "UNKNOWN";
    }
}
 
int main() {
    HANDLE hProcess = GetCurrentProcess();
    HANDLE hThread = GetCurrentThread();
    
    printf("═══════════════════════════════════════════════════════════════\n");
    printf("Windows Priority Demonstration\n");
    printf("═══════════════════════════════════════════════════════════════\n");
    
    // ─────────────────────────────────────────────────────────────────
    // Display current priority settings
    // ─────────────────────────────────────────────────────────────────
    DWORD priorityClass = GetPriorityClass(hProcess);
    printf("\nInitial Process Priority Class: %s\n", 
           PriorityClassName(priorityClass));
    PrintPriorityInfo(hThread, "Initial Thread State");
    
    // ─────────────────────────────────────────────────────────────────
    // Modify process priority class
    // ─────────────────────────────────────────────────────────────────
    printf("\n─ Setting process to ABOVE_NORMAL_PRIORITY_CLASS ─\n");
    if (!SetPriorityClass(hProcess, ABOVE_NORMAL_PRIORITY_CLASS)) {
        printf("SetPriorityClass failed: %lu\n", GetLastError());
    } else {
        printf("Process Priority Class: %s\n",
               PriorityClassName(GetPriorityClass(hProcess)));
        PrintPriorityInfo(hThread, "After Process Priority Change");
    }
    
    // ─────────────────────────────────────────────────────────────────
    // Modify thread priority within class
    // ─────────────────────────────────────────────────────────────────
    printf("\n─ Setting thread to THREAD_PRIORITY_HIGHEST ─\n");
    if (!SetThreadPriority(hThread, THREAD_PRIORITY_HIGHEST)) {
        printf("SetThreadPriority failed: %lu\n", GetLastError());
    } else {
        PrintPriorityInfo(hThread, "After Thread Priority Change");
    }
    
    // ─────────────────────────────────────────────────────────────────
    // Disable priority boosting
    // ─────────────────────────────────────────────────────────────────
    printf("\n─ Disabling priority boosting ─\n");
    if (!SetThreadPriorityBoost(hThread, TRUE)) {  // TRUE = disable boost
        printf("SetThreadPriorityBoost failed: %lu\n", GetLastError());
    } else {
        PrintPriorityInfo(hThread, "With Boosting Disabled");
    }
    
    // ─────────────────────────────────────────────────────────────────
    // Demonstrate MMCSS for multimedia applications
    // ─────────────────────────────────────────────────────────────────
    printf("\n─ Registering with Multimedia Class Scheduler Service ─\n");
    DWORD taskIndex = 0;
    HANDLE hTask = AvSetMmThreadCharacteristics(TEXT("Pro Audio"), &taskIndex);
    
    if (hTask == NULL) {
        printf("AvSetMmThreadCharacteristics failed: %lu\n", GetLastError());
        printf("(This may require the MMCSS service to be running)\n");
    } else {
        printf("Successfully registered as 'Pro Audio' thread\n");
        printf("Task Index: %lu\n", taskIndex);
        PrintPriorityInfo(hThread, "With MMCSS Registration");
        
        // MMCSS-registered threads can request priority changes
        if (AvSetMmThreadPriority(hTask, AVRT_PRIORITY_CRITICAL)) {
            printf("Set MMCSS priority to CRITICAL\n");
            PrintPriorityInfo(hThread, "With MMCSS Critical Priority");
        }
        
        // Unregister when done
        AvRevertMmThreadCharacteristics(hTask);
        printf("\nMCSS registration reverted\n");
    }
    
    // ─────────────────────────────────────────────────────────────────
    // Demonstrate real-time priority (requires elevation)
    // ─────────────────────────────────────────────────────────────────
    printf("\n─ Attempting REALTIME_PRIORITY_CLASS ─\n");
    if (!SetPriorityClass(hProcess, REALTIME_PRIORITY_CLASS)) {
        DWORD error = GetLastError();
        printf("SetPriorityClass(REALTIME) failed: %lu\n", error);
        if (error == ERROR_ACCESS_DENIED) {
            printf("(Requires administrator privileges)\n");
        }
    } else {
        printf("Process Priority Class: %s\n",
               PriorityClassName(GetPriorityClass(hProcess)));
        printf("WARNING: REALTIME_PRIORITY_CLASS can destabilize the system!\n");
        
        // Immediately step back from realtime
        SetPriorityClass(hProcess, NORMAL_PRIORITY_CLASS);
        printf("Reverted to NORMAL_PRIORITY_CLASS\n");
    }
    
    return 0;
}

MMCSS for Audio/Video Applications

Priority Assignment Best Practices

Priority Assignment Guidelines

•Default to Default Priority — Unless you have a specific, measurable reason to change priorities, use the system defaults. Priority elevation should be a last resort, not a first instinct.
•Raise Priority Sparingly, Lower Liberally — It's almost always safe to lower your process's priority (increase nice value). Raising priority risks starving other processes and should require explicit justification.
•Prefer Time Limits to Priority Boosts — Instead of raising priority to complete work faster, consider whether the work can be structured with reasonable time budgets at default priority.
•Match Priority to Temporal Requirements — Use high priority only for latency-sensitive work with genuine real-time constraints (audio playback, control systems). Throughput-oriented work (compilation, data processing) rarely needs elevated priority.
•Avoid Priority Starvation by Design — If using static priorities, ensure lower-priority processes have guaranteed execution windows. Use aging, deadline scheduling, or priority ceiling protocols.
•Consider Priority Inheritance in Lock Design — When high-priority threads may wait for locks held by low-priority threads, implement priority inheritance to avoid unbounded priority inversion.
•Test Under Contention — Priority-related bugs often only manifest under heavy system load. Test with realistic workloads that create scheduler pressure.
•Document Priority Decisions — Priority assignments are often made once and forgotten. Document why specific priorities were chosen so future maintainers understand the constraints.

Don't

Summary: Priority Assignment Fundamentals

We have explored the foundational concepts of priority assignment in operating systems. The key takeaways form the basis for understanding the more advanced topics in the remainder of this module:

Key Takeaways

•Priority semantics vary by OS — Different operating systems use opposite numerical conventions. Always verify whether higher numbers mean higher or lower priority.
•Static priorities offer predictability — Fixed priorities, once assigned, never change. This is essential for real-time systems but risks starvation.
•Dynamic priorities offer adaptability — Priorities that adjust based on runtime behavior improve fairness and responsiveness but sacrifice determinism.
•Priority sources are internal and external — The OS derives priorities from observed behavior (internal) and from user/application specifications (external).
•Linux uses a 140-level hybrid system — Combining real-time scheduling policies (0-99) with normal scheduling (nice values mapping to 100-139).
•Windows uses priority class + thread priority — The combination produces a base priority, which the system dynamically boosts based on events.
•Priority assignment requires discipline — Elevating priority should be a last resort with documented justification. Default priorities are usually correct.

What's Next:

Page Complete

1 / 5