Previous | Table of Contents | Next |
Some systems management products that make extensive use of intelligent agents include Computer Associates CA-Unicenter, Hewlett-Packard Co.s OperationsCenter, and Saber Software Corp.s LAN Management System (a suite of systems management products). The agent software is written for each vendors system platform. The extent to which agent software works with other hardware platforms differs according to vendor. Hewlett-Packards agent software, for example, works on the following UNIX platforms:
Other companies that have awakened to the management possibilities offered by intelligent agents include Apple Computer, BMC Software, Candle Corp., Ki Networks, and Sun Microsystems. IBM Corp. has announced that it will revamp its entire family of management products with a multivendor intelligent agent architecture. The companys Common Agent Architecture, being implemented in phases, uses the DMTF DMI technology, which lets systems administrators gather a variety of desktop system data, such as disk usage and PC configuration, and monitor the performance and status of individual applications. APIs let management consoles, such as HPOV and IBMs NetView for AIE platforms, share event and topology information.
The term agent implies interoperability with other vendors management software, at least on a basic level. This interoperability usually takes the form of SNMP support. The expectation is that if agents are sitting out there on various stations, multiple managers from different vendors will be able communicate with them. But in the distributed environment, the applications are not designed to interoperate with other vendors applications. So what appears as agent technology might really be just a piece of a distributed application.
However, some agents are interoperable. IBMs Systems Monitor/6000, for example, is a true intelligent agent. It provides a MIB interface to UNIX information by exposing UNIX kernel information in SNMP format. Still, products that offer such interoperability are few and far between.
The issue of agent management also needs to be addressed. A large network can have thousands of agents that need to be monitored to find out what agents are actually in the network as well as what they are doing and what has changed. One tool that does monitor agent activity is Legent Corp.s AgentWorks, which is available for several UNIX platforms, as well as for Windows NT and OS/2.
AgentWorks uses icons to represent nodes on the network and allows an administrator to view the overall system, an individual site, a particular network segment, or a particular node. If the systems administrator is looking at an overall view and an error has occurred in a database process, the node where the error occurred appears in red. As the user refines the view to just the node, the database manager agent uses the color red to further pinpoint the problem. If an individual forgets to restart a process, such as an interrupted backup, AgentWorks automatically displays the error.
In addition to allowing systems administrators to create their own intelligent agents, AgentWorks can run as a standalone network management system or interoperate with other management platforms such as SunConnects SNM, HPOV, and IBMs NetView for AIE. AgentWorks is also one of the few products that integrates current SNMP platforms and intelligent agents in a distributed client/server architecture. AgentWorks even allows systems administrators to choose among agent modules for systems, network, application, and database management. Within the basic modules, systems administrators can gradually add management functions as the need arises.
The concept of modular agents is not without its share of potential drawbacks. Because it is possible, for example, to have multiple agents in the same system, more than one agent might respond to a command from the management console. In addition, agents may contend for the same output port when attempting to respond.
This issue is addressed by designating one agent the master and the others as subordinates. The master agent reconciles requests and responses and multiplexes data from subordinates over the output port. However, this scheme introduces a new level of complexity into multivendor environments. Because the masters and subordinates of different vendors are not compatible, they will not work together. Although many vendors have been coalescing around a few approaches for defining master-subordinate agents, such as SNMP Research Inc.s Emanate extensible agent architecture, many users are still forced to buy agents from the same vendor from which they bought the management system.
DMI was designed to promote multivendor interoperability by addressing standardization of the agent interface. To date, however, one of the biggest problems with DMI has been the lack of guidelines on how to retrieve information remotely. The Desktop Management Task Force recently completed its remote specification, which should make it easier to add management capabilities to computers, software, and peripheral devices.
Intelligent agents have a myriad of other uses. BeyondMail from Banyan Systems, Inc. is an example of agent-based messaging. BeyondMails intelligent agents filter, sort, and prioritize mail by sender or topic on behalf of users. A number of sophisticated messaging products have become endowed with agents to help users manage and access data. For example, intelligent agents are used on AT&Ts wireless Personalink network to perform a variety of tasks for mobile users, such as sorting and directing E-mail according to priorities. Among other things, important messages can be sent to the users PDA, and routine messages can be output to a fax machine back at the office.
Intelligent agents offer several compelling advantages for systems management. They detect exceptional events and trigger automated actions to correct problems before they affect systems services, thereby increasing systems availability and reliability. An intelligent agent architecture obviates the need for expensive network polling, because only critical alerts are sent to the management console.
Finally, intelligent agents can be used to automate routine, repetitive tasks, relieving the systems administrator of a heavy work burden. This reduces the overall cost of systems management by speeding problem resolution and permitting businesses to get the most out of their technology investments. This ability to carry out corrective actions on an automated basis, and to keep business systems and applications up and running, can give a company a competitive edge in an economy where information availability and the speed of delivery are critical.
Previous | Table of Contents | Next |