Information Technology Reference
In-Depth Information
Fig. 1.
Diagram of the datacenter model
The
disturbances
are variables that affect the datacenter behavior but we have
no control over, in our case the external temperature
ʱ
k
(
o
C
).
3.2 Scheduling Model
The scheduling model considers a set of tasks to be executed in the datacenter.
Each task is defined by its arrival time, the estimated execution time, and a
deadline, which represents the time by which the user wants the task to finish.
We assume that tasks arrive following a Poisson distribution (parameter
ʻ
=
instance size/simulation time). Tasks durations follow a normal distribution,
with average and standard deviations following a typical cluster, according to
traces from Cluster FING [23] and the Parallel Workload Archive repository.
Three levels are used for deadlines, modeling different SLA between the provider
and the users, according to three slack factors (
sf
):
tight
(
sf <
10%),
medium
(10%
<sf <
30%), and
loose
(
sf >
30%), which represent the time the user admits
the tasks end (over the task duration and since the arrival time).
We apply a dynamic scheduling strategy applying specific QoS and energy-
aware scheduling heuristics, which are described in the following section.
3.3 The Optimization Problem
We want to control the datacenter so that its total power demand
P
k
and temper-
ature
T
k
follow as closely as possible a desired reference demand and temperature
profiles
R
k
and
T
ref
, while minimizing the impact on QoS.
The datacenter executes
n
tasks in the simulation period (
K
steps). Each user
submission requests execution before a deadline
D
(
i
)fortask
i
, and the scheduler
executes the task according to the availability of computing resources and energy
consumption. Each task
i
finishes at time
FT
(
i
), and QoS is evaluated according
the deadline satisfaction. Formally, we want to
minimize
the deviation regarding
the reference power profile (1) and regarding the temperature (2), and the total
time of deadline violations (3) during the simulating period.
