Linux Kernel Architecture

Chapter 2: Process Management and Scheduling

/* 5 */ 335, 272, 215, 172, 137,

/* 10 */ 110, 87, 70, 56, 45,

/* 15 */ 36, 29, 23, 18, 15,

};

The array contains one entry for each nice level in the range [0, 39] as used by the kernel. The multiplier
between the entries is 1.25. To see why this is required, consider the following example. Two processes
AandBrun at nice level 0, so each one gets the same share of the CPU, namely, 50 percent. The weight
for a nice 0 task is 1,024 as can be deduced from the table. The share for each task is 10241024 + 1024 = 0 .5, that
is, 50 percent as expected.

If taskBis re-niced by one priority level, it is supposed to get 10 percent less CPU share. In other words,
this means thatAwill get 55 percent and B will get 45 percent of the total CPU time. Increasing the
priority by 1 leads to a decrease of its weight, which is then 1, 024/ 1. 25 ≈820. The CPU shareAwill
get now is therefore 10241024 + 820 ≈ 0 .55, whereasBwill have 1024820 + 820 ≈ 0 .45 — a 10 percent difference as
required.

0

20000

40000

60000

80000

100000

120000

140000

160000

180000

80 90 100 110 120 130 140

Weight

Priority

Regular task

Realtime task

0

200

400

600

800

1000

1200

120 125 130 135 140

10

100

1000

10000

100000

80 90100110120130140

Figure 2-15: Relation between static priority and load for regular and real-time processes.

The code that performs the conversion also needs to account for real-time tasks. These will get double
of the weight of a normal task.SCHED_IDLEtasks, on the other hand, will always receive a very small
weight:

kernel/sched.c

#define WEIGHT_IDLEPRIO 2

#define WMULT_IDLEPRIO (1 << 31)