Moving Toward Self-awareness

March 7, 2005 (Computerworld)

In the beginning, there was management and security. By today's standards, they were relatively unsophisticated -- the management of a host system and the security of its users. Then came the network. It had to be monitored and controlled to orchestrate all the devices sending packets around. The network also introduced the concept of perimeter security to make sure only authorized users could access the trusted resources.
From these humble beginnings, security and management have spiraled into layer upon layer of complexity as we accelerate toward the adaptive enterprise vision. The Information Age is all about always-on, on-demand resources and utility computing. Consider the crowded management milieu today. We have, to name a few, the management of services, infrastructure, storage, configurations, data centers, traffic, applications, devices of all types, content and compliance, as well as security and identities.
Security and identity management crosses into the security zone. Here the pace of innovation is driven wildly by malfeasance, misfeasance and nonfeasance. The issues that matter here are security for users, storage, configurations, data, applications, devices of all types and content -- the list starts to resemble the one for management.
Much is being written about how the overlap is blurring the historic divide between management and security. Configurations are now a chicken-and-egg dialectic of tight management for strong security versus tight security for strong management. Software patches must be automatically distributed, installed and tested to buttress security on millions of devices around the globe. Network traffic must be monitored to thwart debilitating denial-of-service attacks.
There's no end in sight as autonomic computing gains momentum. But getting to this goal of the on-demand, always-on information infrastructure is going to require new capabilities beyond today's relatively crude management and security tools.
The tools we use for management and security are mostly empirical. We watch and accumulate information so we can react to or, preferably, anticipate situations. The sheer volume of data that must be monitored across the autonomic infrastructure will render empirical management impractical.
Smart vendors, seeking competitive advantage, are adding modeling capabilities to put more intelligence into the adaptive infrastructure. Modeling is the key ingredient for an intelligent infrastructure capable of quickly scaling to meet the demands of change.
These first tools mostly bring deductive reasoning for "what-if" modeling to support the first generation of adaptive infrastructure. What-if modeling predicts the impact of changes to some discrete element buried inside a layer of the information infrastructure. Sophisticated deductive modeling considers the result of changes in two or even three variables. Real-world applications of this have brought us relatively well-behaved systems, albeit ones with fewer variables than the adaptive infrastructure, such as those capable of planning peak water usage during the Super Bowl or managing telephone networks on Mother's Day.
Deductive modeling tools are an intermediate step. They aren't sufficiently powerful or sophisticated to make multivariant decisions across all the myriad elements and layers in the always-on, on-demand infrastructure. As we increase our understanding of the interactions and dependencies, we will acquire the knowledge to do inductive modeling, the more elegant solution.
Inductive modeling goes by the street name predictive modeling and may be the real deal for managing and securing all those elements and layers in the information utility. Predictive modeling is already a hot market for predicting multivariant events in health care, business, air traffic control and meteorology.
Predictive modeling requires a good understanding of the problem and a lot of data, which we're acquiring now for the information utility. The big difference is that predictive modeling starts with the desired outcome. You plug in the numbers of users you must support under a given application workload, and the model returns the optimal configurations for all the elements in all the layers of the infrastructure.
There's no guessing on how storage throughput impacts application performance. The system has the historical data and intelligence to model the impact of cross-domain security and to make adjustments. Neural networks and artificial intelligence agents can monitor critical elements and learn from changes. They train themselves to make rational decisions when faced with complex situations involving many variables.
The predictive infrastructure is "self-aware," a term made famous in the Terminator movies. To make the technology work, we humans must give up the notion of control, for better or worse.