Autoscaling in behalf of monitoring virtual machines

This guide shows the logical steps how we implemented auto-scaling functionalities in a private cloud at 52°North Initiative for Geospatial Open Source Software GmbH.

Auto-Scaling in principle:
The CPU utilization of running instances is requested in defined intervals. If a specific threshold is reached, a new instance will be started recpectively stopped.
This is done by running a web service on each instance that responses the instance's CPU utilzation as well as a program that calculates the average of all CPU utilizations and starts/terminates instances.
Since we are using sysstat for getting the instances' CPU utilization, using Linux as the images' OS (in our case Ubuntu 9.10) is requires .

Following the steps to implement auto-scaling functionalities in a private cloud:

1. Configuring the image:

1. Install sysstat on a Linux image.
   On Ubuntu this is done with the command: sudo apt-get install sysstat
   Furthermore install a servlet container to deploy a webapp, which gives you the instances CPU utilization (described below). (At 52°North we are using Apache Tomcat. This implies that we also need to install a Java Development Kit.)
2. Write a web service that can be requested via HTTP GET with the following logic:

   If this web service is requested, the instance's CPU utilization is retrieved by the console command sar 1 1 After parsing the command's output, the web service responds the CPU utilization.
   This web service is deployed as a webapp in the servlet container as described above.

3. Register the image at Eucalyptus.

2. Implementing auto-scaling functionalities:
To implement the auto-scaling functionalities another program needs to be written.

This program runs on the CC and requests the IP adresses of the running instances by the command euca-describe-instances For these instances the web services, which gives you the CPU utilization (see poin 1), are requested (e.g. for every running instance: "http://instance-ip/cpuLoad"). If this average of all instance's CPU utilizations exceeds respectively goes below a specific threshold, a new instance will be started (euca-run-instances) recpectively stopped (euca-terminate-instances). This is repeated in a defined interval (e.g. every 30 seconds).

That's it!