Information Technology Reference
In-Depth Information
-
y
j,c
is capacity resource of type
c
(e.g. CPU core, computing power, memory)
of the physical machine
).
Based on the above observation, we propose our energy-aware algorithms as
presented in the next section.
M
j
(
j
=1
,
2
, ..., m
4
EPOBF Algorithm for Virtual Machine Allocation
Inspired by the Green500 lists idea [
21
] that uses a metric of performance-per-
watt (FLOPS/watt) to rank energy eciency of HPC systems, we raise ques-
tions: how can we use a similar metric (e.g. TotalMIPS/Watt) as a criterion for
selecting a host to assign a new VM, and is total energy consumption of the
whole system minimum?
We assume that if a host has more number of cores, then it will have more
number of MIPS/Watt. The number of MIPS/Watt of a host is a ratio of total
maximum of MIPS, which is sum of total maximum of MIPS of all hosts cores,
to its maximum power consumption (
P
max
). The objective of our proposed work
is energy eciency. In this paper, we propose our energy-aware scheduling algo-
rithm that denoted as the EPOBF-ST and EPOBF-FT. The EPOBF-ST and
EPOBF-FT are best-fit decreasing heuristics to allocate a new VM to a physical
server in such a way that has the maximum ratio between total computing power
(in MIPS) and power consumption (in Watts).
EPOBF-ST and EPOBF-FT
: The EPOBF-ST (EPOBF-FT) will sort list of
VMs and order them by earliest start time first (earliest finishing time first),
then the core EPOBF will assign a VM to a host that has the highest G value.
For all
]) can be calculated as a
ratio of total maximum of MIPS of the host h (sum of total MIPS of all of cores)
to maximum power consumption at 100% CPU utilization (
h ∈ M
,the
G
h
(denoted in pseudo-code is
G
[
h
P
max
) of the host
h. We called G the metric of performance-per-watt. In summary, the EPOBF
Algorithm 1.
EPOBF-ST and EPOBF-FT: Heuristics for energy-aware VM
allocation in HPC Clouds
1:
function
EPOBF-ST
2:
Input:
vmList - a list of virtual machines to be scheduled
3:
Input:
hostList - a list of physical servers
4:
Output:
mapping (a feasible schedule) or null
5:
vmList = sortVmListByStartTime( vmList )
1
6: return EPOBF( vmList, hostList )
7:
end function
8:
function
EPOBF-FT
9:
Input:
vmList - a list of virtual machines to be scheduled
10:
Input:
hostList - a list of physical servers
11:
Output:
mapping (a feasible schedule) or null
2
12:
vmList = sortVmListByFinishTime( vmList )
13: return EPOBF( vmList, hostList )
14:
end function
Search WWH ::
Custom Search