Linux CPU Scheduling Explained: Context Switches, Run Queues & Monitoring

CPU Scheduler Resources

CPU scheduler is responsible for scheduling two kinds of resources: threads and interrupts. Priority order high to low: 1) Interrupts – devices notify the kernel when they have completed (e.g., network card finished a packet or disk completed an I/O request). 2) Kernel processes. 3) User processes.

1. Context Switches

Most processors can run only one process at a time; on multi‑core systems Linux treats each core as an independent processor. A kernel can run 50–50,000 processes simultaneously. If there is only one CPU, the kernel must balance these processes. Each thread receives a time slice; when the slice expires or a higher‑priority thread preempts it, the thread is placed back into the CPU queue. Switching a thread is a context switch. Each context switch consumes resources to save CPU registers and enqueue the thread. More context switches increase the kernel’s scheduling overhead.

2. Run Queue

Each CPU has its own run queue. Threads are either in a sleeping state (blocked waiting for I/O) or running. A longer run queue means longer wait time for a thread. Load describes the run queue: load = number of running threads + threads in the run queue. For example, on a 2‑CPU system with 2 threads executing and 4 threads in the run queue, load = 2 + 4 = 6.

3. CPU Utilization

CPU utilization consists of:

User time : time spent in user space.

System time : time spent in kernel space.

Wait I/O : time spent waiting for I/O (blocked).

Idle : idle time.

4. CPU Performance Monitoring

Typical acceptable values:

Run queues per processor ≤ 3

User time 56%–70%

System time 30%–35%

Idle time 0%–5%

Context switches correlate with CPU utilization.

5. Monitoring Tools

Common tools: vmstat, mpstat.

6. Performance Troubleshooting Examples

Example 1

High interrupt count, low context switches indicate a single process accessing hardware. User time consumes 85% and few context switches suggest a single application causing the load.

Example 2

Context switches exceed interrupts, indicating excessive time spent switching. High wait I/O shows many threads waiting for I/O, suggesting many threads need to be switched out.

Example 3

CPU0 and CPU1 are handling CPU‑intensive processes, CPU2 is idle, CPU3 handles kernel and other system functions. Commands like ps -psr show which processes run on which cores, and ps -eo pid,ni,pri,pcpu,psr,comm provides detailed process info.

7. Summary

1) Each core’s run queue should be ≤ 3. 2) CPU utilization: roughly 70% user space, 30% kernel space. 3) Excessive kernel‑space usage often indicates heavy load and the kernel spending too much resource on priority scheduling. 4) CPU‑bound processes are penalized while I/O‑bound processes are rewarded.