Introduction

This document is a first sketch at what the requirements are for the "Open Agent". There are lots of agent technologies out there and this document should serve has a guide when evaluating existing agents. The last thing we want to do is create an agent technology from scratch.

Definitions:

Manager - software that is providing a centralized monitoring or management function that interacts with the "Agent" to perform its functionality.

Agent - the shell program that can internally run "Plugins" and has an external interface to enable standards based communication with management systems.

Plugin - program or code that runs within the agent and is used to augment its functionality.

Must have

1. broad platform support - all unix and windows

2. automated agent shell upgrade mechanism

3. standards based remote API

1. SOAP

2. XML-RPC

3. WS-Management

4. WBEM / CIM-XML

4. secure communication mechanism

5. easy mechanism for plugin implementation

6. internal API for plugin to shell communications

7. ability implement plugin (with API access) in major languages:

• Perl, Python, Ruby, C, Java

Should have

1. built into the OS

2. agent deployment mechanism

3. scheduling mechanism

4. plugin deployment mechanism

5. ability to cache data locally (during network failure)

6. ability to initiate connection from agent->manager or manager->agent

Nice to have

1. lightweight foot print

• Plugins compiled when agent is started?

2. standard backend data model or other normalized view

1. CIM?

3. common pre-built plugins like:

1. filesystem, process, log file monitoring

2. create user, start/stop service

3. inventory hardware and software

4. Discovery : try to detect which plugin can be used on the system regarding available ones