maximum temperature, for any cool-
ing parameter a device may have.
sleep states: Irani et al. 23 investigate
an extended problem setting where a
variable-speed processor may be tran-
sitioned into a sleep state. In the sleep
state, the energy consumption is 0
while in the active state even at speed
0 some non-negative amount of energy
is consumed. Hence, Irani et al. 23 com-
bine speed scaling with power-down
mechanisms. In the standard setting
without sleep state, algorithms tend to
use low speed levels subject to release
time and deadline constraints. In con-
trast, in the setting with sleep state
it can be beneficial to speed up a job
so as to generate idle times in which
the processor can be transitioned to
the sleep mode. Irani et al. 23 develop
online and offline algorithms for this
extended setting. Baptiste et al. 11 and
Demaine et al. 21 also study scheduling
problems where a processor may be
set asleep, albeit in a setting without
speed scaling.
minimizing Response time
A classical objective in scheduling is
the minimization of response times.
A user releasing a task to a system
would like to receive feedback, say the
result of a computation, as quickly
as possible. User satisfaction often
depends on how fast a device reacts.
Unfortunately, response time minimi-
zation and energy minimization are
contradicting objectives. To achieve
fast response times, a system must usu-
ally use high processor speeds, which
lead to high energy consumption. On
the other hand, to save energy, low
speeds should be used, which result in
high response times. Hence, one has to
find ways to integrate both objectives.