Information Technology Reference
In-Depth Information
metering), managing energy sources and workloads to minimize electricity costs.
This approach is evaluated on Parasol, a solar-powered micro-datacenter [8].
An alternative to shifting load is to trade-off energy and QoS. The scheduler by
Kriukov
et al.
[15] follows wind generation, using less power by delaying requests
with slack. Aikema
et al.
[2] presented simulation results for a datacenter in the
New York ancillary service market. A selective approach is applied in [7], where
participation as an ancillary services provider is determined by expected profits
(the compensation is weighed against the SLA penalties).
Ghamkhari and Mohsenian-Rad [21] schedule non-critical jobs based on pre-
dicted power output and datacenter load. This is to the best of our knowledge
the only other existing approach for energy-aware scheduling in data centres that
considers air conditioning power. The main differences with our control approach
are: (a) we look at fine time granularity (minutes), broadening the applications of
datacenter energy management,
e.g.
by enabling ancillary services; (b) we apply
a multiobjective approach to the problem, thus considering a range of trade-off
solutions between power, temperature, and QoS; and (c) different schedulers are
studied in order to provide diverse power/QoS trade-offs.
Energy aware scheduling
. A simple optimization approach for energy-aware
scheduling assumes energy and performance as independent. A more compre-
hensive one is to optimize performance and energy at the same time, modeling
the problem as a multiobjective optimization. Algorithms are oriented to find
Pareto optimal schedules;
i.e.
, no scheduling decision can strictly dominate the
others with better performance and lower energy consumption at the same time.
Khan and Ahmad [12] applied game theory for scheduling independent jobs,
simultaneously minimizing makespan and energy on a Dynamic Voltage Scal-
ing (DVS)-enabled grid system. Lee and Zomaya [17] studied several heuristics
to minimize the weighted sum of makespan and energy using a makespan con-
servative local search to slightly modify scheduling decisions when they do not
increase energy consumption, in order to escape from local optima. Later, Mez-
maz
et al.
[22] proposed a parallel bi-objective hybrid genetic algorithm (GA)
for the same problem. significantly reducing the scheduler execution time.
Kim
et al.
[13] studied the deadline constrained scheduling problem in ad-hoc
grids with limited-charge batteries, proposing a resource manager to exploit the
task heterogeneity while managing energy. Li
et al.
[19] introduced an online dy-
namic strategy with multiple power-saving states to reduce energy consumption
of scheduling algorithms. Pinel
et al.
[26] proposed a double minimization ap-
proach for scheduling independent tasks on grids, using an heuristic to optimize
makespan, and then a local search to minimize energy consumption. Lindberg
et al.
[20] proposed six greedy algorithms and two GAs to solve the makespan-
energy scheduling problem subject to deadline and memory requirements.
Le
et al.
[16] proposed a scheduler for deciding the datacenter to run virtual
machine requests and job migration, taking into account electricity price and
temperature, which can trigger AC activation, increasing power consumption.
Our previous work [24] introduced an energy consumption model for multi-
core computing systems based on the energy the system requires to operate at
