Information Technology Reference
In-Depth Information
Each workload pattern above represents a typical scenario. For example, workloads
of Application 1 and Application 3 slightly differ from each other but show a very
common pattern that fluctuates consistently and reaches peaks in the middle of the
day. In addition, workload of Application 2 shows a very steady pattern with several
wiggles.
Application 1
Application 2
Application 3
Fig. 4.
Workload Patterns of Application 1, 2 and 3
The fixed parameters of the experiment are shown in Table 1. Each workload is split
into 144 intervals with 10 minutes per interval. Each application is initialized and
deployed with 1 instance. Coefficients shown in Table 1 are obtained based on expe-
riences. The values of these coefficients could be determined by using PID tuning
techniques [14].
Table 1.
Fixed Experimental Parameters
3
Number of Applications
Minutes per Interval
10
Number of Intervals
144
Initial Instances of each Application
1
Coefficient
2
Coefficient
1.5
Coefficient
1
Coefficient
0.5
Coefficient
1
5.2
Evaluation of Resource Management Framework
Besides the fixed experimental parameters shown in Table 1, there are also three
kinds of tunable experimental parameters including Set Point of Missed Deadline
Ratio for each application (denoted as MissRatio_s), Set Point of CPU Utilization for
each application (denoted as U_(cpu_s )) and VM Safe Threshold. For the clarifica-
tion and simplicity of the following analysis, MissRatio_s and U_(cpu_s) for all
applications are set together. The proposed resource management framework is eva-
luated under different settings of tunable experimental parameters.
