Mailbag: Readers respond to agent overload, Part 1

* Readers say yes to agents

The idea of software agents installed on desktops and servers across large enterprise networks gets people talking. An article and a newsletter I wrote a couple of weeks ago generated a lot of feedback from my readers.

And I must say, the responses are mixed, many can make the argument for and against agents. Some say they are a necessary evil and others say they are just plain evil. Here I will share the comments in favor of software agents.

James Kritcher, vice president of IT at White Electronic Designs in Phoenix, says agents have their place and while the idea of agent-less tools appeals to him, the technology is not where he needs it to be yet.

"I think that while agent-less products are getting better -- and are taking advantage of built-in functionality such as WMI, agents will always be required. In many cases, agent-based tools provide richer functionality and better performance. As far as security, it depends on the application and requirements. Agent components may be susceptible to user disablement. On the other hand, some agents continue to work and spool information in the event of a network failure -- and upload the monitoring results when service is restored. Agent-based systems provide their own authentication methods. With the agent running as a service, the server authenticates using a proprietary method that doesn’t require a user name and password on the client. Agent-less systems generally must use a pre-existing privileged user account. If you are looking for rogue network devices -- such as routers, wireless access points, etc. -- then an agent-less system is mandatory -- obviously an unauthorized device is not going to have an agent installed. Personally, I prefer agent-less products. In addition to the overhead of installing, managing, and running agents, we prefer to run critical servers as clean as possible, minimizing the number of services and applications which could potentially cause issues. I’m not sure that you can entirely avoid the use of agents. While agent-less offerings are improving, they still tend to be best for high-level observation. If advanced functionality or granular control is required, agent-based systems still have the edge."

Kim Jahnz, lead systems management analyst, at Aurora Health Care in Milwaukee, says vendors she works with make adopting agents a more appealing option. And like Kritcher, Jahnz thinks avoiding agents is not a realistic option for the time-being, considering what's available from vendors today.

"If you want any kind of performance monitoring other than the very basic, then an agent is required. As far as troubleshooting, it depends completely on the issue that the client is having, whether we feel that agents would help with data gathering, or helping to pinpoint the issue. I prefer super agents for servers. I am very pleased that the new CA agents will have all of the architecture so that I can, with one agent, monitor what I need. This is great; it should be less expensive and the maintenance should be a lot less too. Ideally I would like to be able to load the framework agent, and then pick and choose what 'modules' to load into that framework. I believe this is the CA vision. The key is that the 'modules' need to be all-encompassing, using application performance monitoring agents as well as system performance. Both should be an option."

To avoid using agents, she says: "You would need to have the agent essentially built-in or embedded into either the operating system or application with a standard way of sending out data and information so that systems like Spectrum could grab it and turn it into alarms, etc. I don't see this as a quick or simple option, since a lot of our monitoring would fall under 'non-standard.' The vendors would have a hard time coming up with every option to meet every need."

One anonymous reader contributed the following:

"As for the question of how many agents, I think a standard would be good, and in that case only one is needed. Even if agent functionality is embedded in the [operating system], there's always something in the environment that is 'not included.' This is what makes every organization's system environment unique. What am I willing to live with on my endpoints?: Definitely a light-weight agent if it is needed; and one that doesn't require constant updates and one that can be centrally managed. Of course, this is what everyone probably wants. It's a good question though because it brings to mind a lot of other issues -- e.g. IBM's autonomic computing model, BMC's agent-less monitoring, etc."

Join the Network World communities on Facebook and LinkedIn to comment on topics that are top of mind.

Copyright © 2007 IDG Communications, Inc.