Understanding Linux load average

We often hear the term “load average” but how much do we know about it? In this post, I will try my best to explain in the fewest letters everything we need to know about load average.

First, how do we determine the load average? We can use commands like uptime and top. Both will present you 3 values.

bash-3.2$ uptime
 03:46:39 up 703 days, 11:00,  2 users,  load average: 3.04, 3.10, 3.08
bash-3.2$

From the output above, 3.04, 3.10, and 3.08 is the load average for the past 1, 5, and 15 minutes respectively. You may also use “cat procloadavg” and will get same result for the 1st 3 columns.

Now, what is LOAD AVERAGE? It is the average of all cpu loads on your system.

What is a CPU LOAD? It is the NUMBER of processes using + NUMBER of processes in queue for a single cpu core. It is NOT the CPU usage.

What is CPU USAGE? It tells us how active is your cpu cores.

To determine CPU LOAD on each cpu core, we follow these 2 formulas:

load per cpu core = (load average) / (# of cpu cores)

load per cpu core = (processes using the cpu core) + (processes inqueue for that cpu core)

where:

load average = the one reported in procloadavg, uptime, top, etc ..

# of cores = grep -c ^proc proccpuinfo

To better illustrate, look at the scenarios below for a 1-core and 2-core machine.

1-core machine (CPU A only):

0 process using CPU A = load average of 0 (under capacity)

1 process using CPU A = load average of 1 (at max capacity)

1 process using CPU A, 1 process waiting in line = load average of 2 (overcapacity)

1 process using CPU A, 2 process waiting in line = load average of 3 (overcapacity)

… and so on ..

2-core machine (CPU A and CPU B):

0 process using CPU A, 0 process using CPU B = load average of 0 (under capacity)

1 process using CPU A, 0 process using CPU B = load average of 0.5 (under capacity)

1 process using CPU A, 1 process using CPU B = load average of 1 (max capacity)

1 process using CPU A, 1 process using CPU B, 1 processes waiting in line = load average of 1.5 (slightly overcapacity)

1 process using CPU A, 1 process using CPU B, 2 processes waiting in line = load average of 2 (overcapacity)

… and so on ..

Based from the examples above, a load average of 1 means all of your cpu cores are at max capacity. Load average of 2 means each cpu core has 1 process running and 1 process waiting in line.

There are instances where your load average is high but the CPU usage is low. An example is when you have several hunged processes occupying all your cpu cores. Sinced hunged, those processes doesn’t generate CPU activity but they still hold the CPU cores. When all CPU cores are being held, no other processes can use them. So if processes can’t get hold of the CPU, this can slow down your system.

In short, a system with LOW CPU USAGE but with HIGH LOAD AVERAGE can still slow down your system.