EPOBF: Energy Efficient Allocation of Virtual Machines in High Performance Computing Cloud - Transactions on Large-Scale Data-and Knowledge-Centered Systems - page 80

Table 1. Server characteristics (Type A: HP ProLiant ML110 G5, Type B: IBM x3250,
Type C: Dell PowerEdge R620)

Server Type A Type B Type C

CPU 1x Xeon 3075 1x Xeon X3470 2x Xeon E5-2660

with 2.66GHz with 2.93GHz with 2.20GHz

Number of cores (1) 2 4 16

Maximum of MIPS/core (2) 2660 2933 2660

Pidle (Unit: Watt) 93.7 41.6 56.1

Pmax (Unit: Watt) (3) 135.0 113.0 263.0

TotalMIPSofallcores 5,320.0 11,732.0 42,560.0

(4) = (1) * (2)
TotalMIPS/Pmax 39.4 103.8 161.8

(5) = (4) / (3)
RAM (GB) 4 8 24

Network bandwidth (Kbits/s) 10,000,000 10,000,000 10,000,000

In the Algorithm 2 , the EPOBF tries to allocate all VMs to physical
machines. In the lines 5-8, it iterates on the list of VMs and then places
the first VM to a physical machine

h whose available resources can satisfy

required resources of the VM and the h

is maximum. It does so by invoking the FindHostForVmByGreenMetric func-
tion. At line 17 , the findCandidateHosts(vm, H) function returns a set of can-
didate hosts which have sucient available resources to satisfy all required
resource constraints on the

G [ h ] value of the physical machine

vm . Lines 18-37, the EPOBF chooses the best host

h candidateHosts

from the set of candidate hosts (denoted as ), in which the

h

host has the maximum ratio of total host MIPS and maximum power host

(denoted as

,
and the GetPower(1) function returns the maximum power consumption of the
host

G h ). The TotalMIPS is the maximum total MIPS of the host h

h .

5 Performance Evaluation

This section presents the results obtained from our comparative simulated
experiments. The simulated experiments aim to demonstrate and evaluate per-
formance of the proposed energy-aware scheduling heuristics which are best-
fit decreasing heuristics (denoted as EPOBF-ST and EPOBF-FT, in com-
parison with (1) PABFD, which is a popular power-aware VM allocation
heuristic, (2) VBP Greedy L1 and L2 that are two norm-based vector bin
packing heuristics. Our proposed EPOBF-ST and EPOBF-FT heuristics are
presented in Section 4 . The following section will present the PABFD and
two norm-based vector bin packing (VBP Greedy L1 and VBP Greedy L2)
heuristics.
PABFD : The PABFD (Power-Aware Best-Fit Decreasing), whose objec-
tive is minimizing total energy consumption of physical machines, uses as a