Overview

The Compute Grid infrastructure provides an implementation of the master-worker pattern using space-based commands. The Compute Grid includes ready-made services, with administrative service UI support.

Master-Worker Pattern

Expand this...

In the master-worker pattern, a job represents specific business logic, comprised out of a set of tasks. A job is responsible for creating tasks and collecting results. A task is the smallest unit of work that needs to be executed by the workers.

External references for master-worker pattern: Artima Developer, "Make Your Compute Server Robust and Scalable" OnJava.com, "Space-Based Programming"

Compute Grid Architecture

The Compute Grid has three main components:

• Master – a master owns a specific job. The master writes tasks using the write() API and waits for a response on those tasks, using the take() API.
• Scheduler – an optional service, used to manage execution of tasks in advance. Tasks are routed to scheduler by the master; the scheduler determines which workers should execute them, based on priority, CPU utilization, etc. The scheduler can be implemented as an external server.
• Workers – workers represent a compute agent. Workers wait for tasks using the take() API. Once a task has been taken, the worker executes the task using the execute() method on the Task Entry interface. The result of the task is written back to the space as a response Entry. A task ID is used for correlating between between task and response.

To define behavior and state, you can create Task Entry objects – this provides a ready-made dynamic Compute-Grid environment.

Dynamic Scaling of Worker Units

Unlike in dynamic caching scenarios, in a Compute Grid scenario the Service Grid uses space backlog to monitor the load on existing workers and machines.

Expand this...

Section Contents

• Creating a Custom Job — How to implement a custom Job, Task, and Result using the relevant classes.
• Deploying Compute Grid and Running Jobs — An example of how to prepare the environment, deploy the Compute Grid service, and run a job.
• Running Jobs from the Command Line — How to write a command line interface for executing jobs.