Green Cloud Scheduler

Contents

1. Overview
2. Behavior
3. Interaction with OpenNebula
4. Deployment
5. Configuration options

Overview

Behavior

1. Reads servers's characteristics by using the Java OCA API HostPool class.
2. Reads the OpenNebula managed virtual machines by using the Java OCA API VirtualMachinePool class.
3. Updates the energy-aware cloud context model.
4. Whenever a new VM is submitted to OpenNebula or an existing VM is deleted (see Figure 1), an OpenNebula Hook event is triggered. GCS will react to the event and will start analyzing the cloud context for finding the optimum virtual machine distribution.

• In the Monitoring phase the GCS context is updated with new cloud information.
• In the Analysis phase, GCS analyzes the context and evaluates the cloud greenness level and decides if new actions are necessary for improving the cloud's energy efficiency.
• In the Planning phase the GCS uses a learning algorithm for generating an adaptation plan consisting of deploying/migrating virtual machines and turning on/off servers actions.
• In the Execution phase, GCS uses the following tools to enforce the adaptation plan defined in the planning phase:
  • OpenNebula Java OCA API for deployment and migration of virtual machines
  • wakeonlan for turning ON servers
  • ssh for turning off servers: ssh hostname sudo /sbin/shutdown –h now

Interaction with OpenNebula

Cloud Monitoring

Retrieving Servers Information

• Maximum CPU frequency retrieval
  For finding the maximum CPU frequency for a server the cpufreq-utils tool is used, due to the fact that JAVA OCA API reports only the current CPU frequency. If cpufreq-utils is not installed on the OpenNebula nodes, the info from OCA is used instead.
• MAC address retrieval
  To retrieve the MAC addresses for each server defined in OpenNebula servers pool, the GCS reads the data from ./config/arpTable and if the MAC for a server is not defined, GCS pings the server, issues an arp and parses the output. The retrieved MAC address is then stored in the ./config/arpTable for later use.

Retrieving Virtual Machines Information

• Virtual Machine Requirements Information
  One issue with the VM information is that the CPU field from the VM template contains only the number of CPUs the virtual machine has requested, while the required CPU frequency is only an optional value in the REQUIREMENTS portion of the VM template (see Figure 2).

  Figure 2: VM template example

  If the REQUIREMENTS portion contains information about the required CPU frequency, this information is collected for futher use. If no CPU frequency info is specified in the virtual machine template, the data from the entry VM_DEFAULT_CPU_FREQUENCY in the /config/config.properties is used instead (See Table 1).
• Virtual Machine Creation/Deletion Information
  The GCS is notified when virtual machines are created/deleted in/from the OpenNebula VM Pool by using OpenNebula hooks mechanism. The OpenNebula hooks are defined in the /etc/one/oned.conf file by entering the lines presented in the Figure 3.

  Figure 3: OpenNebula Hooks Configuration

  Download the GCS hooks configuration /etc/one/oned.conf excerpt here.

Tools used by GCS

• Wakeonlan
  The GCS uses the wakeonlan tool to remotely wake up the data center servers. The following command is used: wakeonlan_tool MAC_Address .
• Cpufreq-utils
  The GCS uses the cpufreq-utils package to retrieve the maximum frequency of the target server .The following command is used: cpufreq-info -l .
• OpenSSH and Shutdown
  The GCS uses the OpenSSH to turn off the data center servers remotely. The following command is used for turning off servers: ssh hostname sudo /sbin/shutdown –h now. To enable the remote execution of /sbin/shutdown, on each server the following configuration steps are required:
  • Execute : sudo chmod u+s /sbin/shutdown
  • Add “oneadmin ALL = NOPASSWD: /sbin/shutdown” in /etc/sudoers to allow oneadmin to use sudo without password on /sbin/shutdown. If the Green Cloud Scheduler is run by an other user, replace oneadmin with the proper username.

Deployment

1. Install wakeonlan on the control node
2. Optionally (but strongly recommended), install cpufreq-utils on the OpenNebula nodes
3. Setup passwordless shutdown for OpenNebula Nodes
• Execute sudo chmod u+s /sbin/shutdown
• Add "oneadmin ALL = NOPASSWD: /sbin/shutdown" in /etc/sudoers to allow oneadmin to execute sudo /sbin/shutdown without password. If you run the OpenNebula client from other user than oneadmin, replace oneadmin with your user in the previous command.
4. Edit the /etc/one/oned.conf file to enter the vm creation and deletion hooks. Restart the one deamon.
5. Install Ant on the OpenNebula controller node. Enter the java JDK location in the ./build.xml file by changing the jdk.home property. Issue ant in the GreenCloudScheduler directory to compile the software.
6. Kill the Default OpenNebula Scheduler process: sudo killall mm_sched
7. Run the GreenCloudScheduler.jar from the directory where it is located : java –jar ./GreenCloudScheduler.jar or use ant ant run

Configuration options