Information Technology Reference
In-Depth Information
Conditions of Simulations
Physical Servers:
120 homogeneous physical servers; each capable of hosting 20
“economy” VMs (see Table 1).
Virtual Machines: As shown in Table 1, VMs has been classified into economy,
standard, advanced, and high performance. These types were determined by
reference to the Amazon EC2 and the Fujitsu Global Cloud Platform.
Virtual Systems:
A virtual system consists of 2-12 VMs that cab be comprised
of various types. All VMs in a virtual system will be deployed or deleted
synchronously on the basis of a request from a user.
Duration of a simulation: 1 year.
Overall CPU utilization:
The average overall CPU utilization in a data center
starts from 0% (no VM deployed) and grows up to 60% by the end of one year.
Note that some deployed virtual systems are deleted during the simulation. In
this preliminary evaluation, VM placement that resulted in an over-committed
state was not allowed.
Electricity costs: We assume that a physical server will consume 300 W of
electricity (e.g., Fujitsu PRIMERGY RX200S5 equipped with two Intel Xeon
2.53GHz processors and 24GB memory). Based on the cost of the special high-
tension voltage power in Tokyo, Japan, the cost is approximately 3.4 yen per
hour.
Table 1.
Types of VMs
Type
CPU
Memory (GB)
Economy
1
1.7
Standard
2
3.4
Advanced
4
7.5
High Performance 8 15.0
Note: “CPU=1” is equivalent CPU performance of Intel Xeon 1.0 GHz.
Simulation Results.
Configurations of the weighted values for the applied plug-in
modules are summarized in Table 2. “A” represents the lowest boundary of electricity
cost. “F” distributes VMs considering both availability and repulsion (i.e., no
electricity saving). The weight of availability and repulsion are gradually increased
from “C” to “E.”
Three patterns of request sequences, p1, p2, and p3, were assessed. Under
conditions described in above, each virtual system was given randomly generated
parameters: type and number of VMs, date and time of deployment, and deletion.
Figure 5 shows the average calculation time to find an optimum VM placement by
the proposed mechanism and a brute-force method (simulated on a PC equipped with
Intel Core i5-2520M 2.50GHz CPU and 4GB memory). The x-axis represents the
number of VMs included in a virtual system. We can see that the proposed
mechanism realized a practical calculation time even the number of VMs was
increased. For example, for 8 VMs, the average calculation time by the proposed
mechanism and a brute-force method were 0.097sec and 160sec, respectively.
