logo
Go to the homepage of the Vrije Universiteit. Goto the homepage of the faculty of sciences. Goto the homepage of the department of computer science.

Cluster and Grid Computing (Practical Work)


Second Assignment:


A Resource Broker Adaptor for the Amazon Cloud :


As a first assignment of the C&CG practical course you have implemented the Joinc task farming library. By using the JavaGAT, you could schedule jobs to run on the compute nodes of the DAS system.

For the second assignment, you have to implement a resource broker for the Amazon Cloud. This resource broker will be implemented as a GAT adaptor (i.e., a plugin for the JavaGAT). Whenever a job is submitted using your resource broker, it will reserve a machine at amazon, and submit your job to that machine.

Do not start from scratch! The easiest way to create this resource broker is to use the Typica library to reserve machines at Amazon. Once machines are available, you can use the existing Trilead SSH GAT adaptor to submit tasks to these machines. Like in the first assignment, a tarball with a code template is provided. For details click on the menu items on the left.


Details:

For this second assignment you are supposed to:

  • Create a resource broker adaptor for the Amazon Cloud.
  • Use the Typica to reserve and release machines on Amazon. Use existing SSH GAT adaptors to deploy tasks to those machines. Also implement a simple scheme that allows you to re-use a machine when a job has finished. Amazon charges a small pre-hour fee for each machine, i.e., if you use a machine for 1 minute, you must pay for a complete hour. Therefore, reusing a machine will save you money!
  • Change your JOINC implementation to use your Amazon adaptor instead of the globus one.

You are expected to hand in the following items:

  • a tarball containing your implementation of the Amazon Resource Broker and modified JOINC implementation.
  • documentation describing what you have done (very short, e.g., 4 pages). Include some application measurements that use your JOINC implementation together with your adaptor to illustrate that your implementation works. For example, you can show a speedup graph for PRIME or TSP for 1 to 8 machines. You can also take a closer look at the cost-effectiveness of your approach (e.g., show the price-per-task for different number of machines).


If you spot a mistake, please e-mail the maintainer of this page.
Your browser does not fully support CSS. This may result in visual artifacts.