Regarding "Deploy on OS" concerns me (in my microcosm of Cisc) as we are very cautios of making target system "unsupported" by installing non-Cisco certified software on devices. As well, as Cisco moves to more appliance like Linux based devices, they do not allow ANY access to the device other than a locked up CLI shell to perform basic functions for the OS.

I still will relay heavily on probes, collectors, or off-box agents accessing target devices remotely to collect data via SNMP, WMI, Perfmon to extract the required data to monitor devices.

This is something I've been thinking about for a while. I don't have much to add at this point. But I will say that I think one of the most important pieces will be a solid plugin/extension API. Really, at it's core, an open agent could be just a very basic core that supports loading of various plugins which enable the particular features needed. That way, one agent could serve the needs of monitoring, deployment, remote control, etc.

I see several issues with yet another agent

• Deployment

How is the agent deployed to the target system ?

Without a centralized software management system (which needs an agent ...), this means manual install on every client. Not very promising.

• Security

Pull or push model ?

With push, the agent needs to open a port and listen to incoming request. Any open port is usually considered a security risk.

Using pull, it must regularly check for management request. One cannot do immediate requests but wait for the next pull. And with lots of clients, pulls will clutter the network.

• Discovery

How can a client (running the agent) be discovered ?

• Protocol

How should the agent talk to the management application ?

REST would be nice (and maybe the only reason for a new agent).

But this still leaves resource representation open.

XML is considered bloated and not very useful for monitoring. Monitoring must be done on the client (RRDtool ?!) as well as policy evaluation. Policy violations (disk full !) can then be reported asynchronously.

• Data Model

How to represent managed entities and their relations ?

It should be object oriented and allow discovery of relations. If a network switch fails, you do want to know which services are affected. The model must support to discover services routed over the failed switch.

• In band - out band

Management of clients in case of critical hardware failures

If the power supply burns, the CPU just overheated or the kernel crashed, the system must still be manageable. This requires as Baseboard Management Controller (BMC) ideally speaking the same language as the agent.

• Reusability

Writing management instrumentation is hard. How can existing instrumentation code being reused ?

• Interoperability

Mixed Windows/Unix/Linux environments are fact today.

Who's going to develop and support the agent across these environments ?

So what's wrong with existing standards and solutions ? CIM and WS-Management have solved most of the above mentioned problems.

Instead of reinventing the wheel, effort is better spend in making these standards more mature and interoperable. Interestingly enough, Microsoft opted for WS-Management because customer asked them for interoperable management.

OMC should jump onto this bandwagon and push Microsoft further down this road.