Power & Non-Power aware

Non-Power aware:

Generate random workloads with randomized constants VMs, Hosts etc., could be achieved with the class non-power aware class of examples.power.random within org.cloudbus.cloudsim. The corresponding class is NonPowerAware. The sequence of classes to be called are:

• RandomConstants
• RandomHelper
• RandomRunner
  • Actual implementation is called from here

Power Aware Planetlab:

Dataset of CPU utilization traces collected from the Planetlab in 10 days in March 2011. Host and VM information collected during these days. 288 entries in a dataset, every 5 mins interval data is logged for 24 hours. ie., 146-179-suf..net_xx (node, VM) as nomenclature for each file

Power Usage Effectiveness

Effectiveness = Power used by DC / Power used by IT equipment

Usually power used by non-IT equipment(cooling, fans) etc., should be far less to call an it an effective data center.

Policies:

Allocate VMs to Hosts and Schedule Cloudlets to Vm. Migrate Vm between Hosts in same or different data centers.

Datacenter simulation

The several ways to simulate a heterogeneous power aware data center based on a combination and allocation and selection policies applied for VM operations. Described below are few of them with a particular real PlanetLab workload. The remaining configuration parameters are in the Constants and PlanetLabConstants classes.

VM Allocation policy

• Iqr: Inter Quartile Range (IQR)
• Lr : Local Regression (LR)
• Lrr: Local Regression Robust (LRR)
• Mad: Median Absolute Deviation (MAD)
• Thr: The Static Threshold (THR)

• VM allocation policy and Maximum Correlation (MC) VM selection policy.

VM Selection policy

Maximum Correlation (MC)

Minimum Utilization (MU)

Minimum Migration Time (MMT)

Random Selection (RS)

A simulation of a heterogeneous power aware data center that only applied DVFS, but no dynamic optimization of the VM allocation. The adjustment of the hosts' power consumption according to their CPU utilization is happening in the PowerDatacenter class.

A simulation of a heterogeneous non-power aware data center: all hosts consume maximum power all the time.

Power Aware Models

Analysis of the cloud simulation environments resulted in questions in several directions. The narrow down made on the two important areas. Models and datasets. There are several was policies that helps in scheduling resources in the cloud. The resources are part of a data center. A broker, the one schedules the simulation by invoking the events implemented from sim entity of the CloudSim toolkit. Hosts and cores as pre-determined in a datacenter. Virtual machines are allocated to hosts such that the cloudlets or tasks or workflows are executed within in. The workflows datasets are derived by executing such applications in a data center setup. Examples include Planetlab workload , TPC-W benchmarks, Google cluster trace.

The scheduling models in CloudSim uses the following VM Allocation polices

Abstract

Migration Abstract

Migration Quartile Range

Migration Inter Quartile Range

Migration Local Regression

Migration Local Regression Robust

Migration Median Absolute Deviation

Migration Static Threshold

Simple

Next is the set of VM Selection policy

Maximum Correlation

Minimum Migration Time

Minimum Utilization

Random Selection

The UtilizationModel interface needs to be implemented in order to provide a fine-grained control over resource usage by a Cloudlet.

The UtilizationModelFull class is a simple model, according to which a Cloudlet always utilize all the available CPU capacity.

The UtilizationModelNull class is a simple model, according to which a Cloudlet always require zero capacity.

The UtilizationModelPlanetLabInMemory class implements PlanetLab data set in memory.

The UtilizationModelStochastic class implements a model, according to which a Cloudlet generates random CPU utilization every time frame.