Information Technology Reference
In-Depth Information
deadline violations. If no hole is available to execute a task, EFTH selects
the machine that provides the minimum finishing time to the task.
5 Experimental Analysis
This section reports the experimental analysis of the proposed MOEA for energy-
aware datacenter control and scheduling for a simulated data center with the
characteristics described above. Both the MOEA and the datacenter simulator
were implemented in MATLAB.
5.1 Problem Instances
Instances are defined by a
workload
,a
scenario
,anda
reference power profile
.
Workloads are sets of tasks. We consider
non-deferrable workloads
and
de-
ferrable workloads
, where 25% of the tasks are allowed to end after the deadline
without having a negative impact in the QoS perceived by the user. We study
three different workload dimensions:
low operation
(50 tasks in 150 time steps),
normal
(75 tasks in 150 time steps), and
full steam
(100 tasks in 150 time steps).
The hardware scenarios assume 64 Atom-based servers in the datacenter, as
in Parasol [8]. The power consumption of each server is 30W(max), 22W(idle),
3W(asleep). We consider a time horizon of 75 minutes (150 30-second time steps)
in the simulation, and an average utilization of 50%, to allow for a reasonable
task planning (utilization values as low as 15-20% have been reported [27]).
We consider three reference power profiles to follow (see Fig. 2, percentages
represent a fraction of the maximum datacenter power):
-
Profile A:
20% during 25 time steps, 80% during 25 time steps and 20%
during 25 time steps. This scenario studies how the system responds to step
changes (both up and down) in power profile.
-
Profile B:
50% during 15 time steps, 80 % during 10 time steps, 20% during
20 time steps, 80% during 10 time steps and 50% during 20 time steps. In
this situation, it is known in advance that demand will have to be dropped in
the near future (
e.g.
forecast indicates that renewable generation will drop)
and we decide increasing power demand before and after the drop.
-
Profile C:
80% during 25 time steps, then a linear ramp decreasing to 20%
during the course of 25 time steps, and then 20% during 25 time steps. This
scenario tests how the control responds to ramp changes, a very common
type of power change in the electricity market.
5.2 Multiobjective Optimization Metrics
In this work, we apply several relevant metrics to evaluate the results obtained by
the studied MOEAs, regarding the goals of converging to and correctly sampling
the set of non-dominated solutions of the problem [5][4]:
