In an industry that loves buzzwords, autonomic computing continues to attract attention. Can the promise of self-managing IT systems ever be met, and how will businesses change if that happens?

Published in Technology & Business, March 2003

In the beginning, it was all about sweat. If your body temperature becomes too high, you will sweat automatically. You don't need to think "Gee, I'd better start perspiring", because the sweating function is controlled by your autonomic nervous system, which is also responsible for many other key bodily functions such as breathing and making sure your heart continues to beat.

IBM research vice president Dr Paul Horn appropriated this concept in late 2001 when he launched IBM's Autonomic Computing Manifesto, which defined an approach to computing based on building, in effect, an autonomic nervous system for IT. Rather than requiring administrators to check them all the time, IT systems should be capable of monitoring and managing themselves, Horn reasoned. What else are we going to use all that processing capacity for?

Since that time, IBM has stuck doggedly to its vision of incorporating autonomic technologies into its products, and many other vendors have also talked up the concept. The notion of autonomic computing seems here to stay, but does it offer real benefits?

Defining autonomic

In its autonomic computing FAQ, IBM defines the concept as follows: "Autonomic computing is an approach to self-managed computing systems with a minimum of human interference."

And why don't we want human interference? Because it costs money. "Although the costs of IT continue to go down, the overall cost of managing the IT infrastructure continues to rise," says Ric Telford, director of autonomic computing architecture at IBM. In other words, it's cheaper to buy stuff these days, but the more you get, the harder it is to control and the more people you need to employ to do that. "You can't have an on-demand business with the current complexity that systems have today," says Telford. "Complexity drives the amount of management associated with an IT infrastructure.

"Forty percent of IT costs today are on staff and administration," he adds. "At least one-third of that is focused on administrative tasks and the basic management of the infrastructure. The ultimate goal of autonomic computing is to allow the IT infrastructure to be managed and controlled by what your business processes are."

Part of the reason for the shift is that system complexity can quickly move beyond the level of individual comprehension. "Simple things such as running out of disk space should be easy to spot, but with so many components to watch, that could be a person's job just to check it," says Graham Ridgway, CEO of software vendor Touchpaper. "The number of parameters to monitor means you either need an automated rules-based system or an army of experts."

Although the autonomic label has caught on, IBM's rivals are quick to point out that the vision isn't entirely fresh. "It's a new name for something that people have been doing for quite a while," says Andy Cooper, marketing manager for information management solutions at Computer Associates (CA). "We've been working towards that goal for about five years." The autonomic vision also has many elements in common with HP's utility-driven view of the universal data centre.

Even the notion of building systems that emulate in some way the biological functions of humans is not unique to IBM. Microsoft, always quick to latch onto a popular vision, has also talked up the concept. "The service components and even the systems themselves have to exhibit more properties that are almost like the biological metaphors where they're self-organising and self-healing. Without that it will be hard to believe that people will be able to keep up with the amount of computing capability that surrounds them in their daily lives," Microsoft chief technical officer of advanced strategies and policies Craig Mundie told a security conference last year.

Going overboard Despite the downplaying of the original bioscience elements and the simplicity of the overall concept, the autonomic vision remains as susceptible to technobabble as any other futuristic vision. "To build a rich-media service grid for automatically managing the life-cycle activities of services, it is essential to have an automatic resource management system to provide a cost-bounded capacity guarantee illusion to services over distributed heterogeneous resources," HP Labs researchers Yong Yan and Bo Shen noted in a recent paper. Answers on the back of a postcard, please.

IBM these days is placing less emphasis on the biological origins of the autonomic vision, effectively using the concept as shorthand for a range of self-managing and self-healing technologies. "I wouldn't say it was a conscious shift," explains Telford. "After a while, once you feel like that message has gotten across, it's not so important to focus on where that idea came from as where you want to go with it."

That point notwithstanding, "I think [bioscience] will clearly be an area to continue to tap into," Telford says. "Part of the initiative is to focus on working with universities and academia -- it isn't an IBM-centric initiative -- we want to drive it, but we want the good ideas to come from all over. That's a rich area for potential advancement of new technology."

Could it possibly work?

For IT managers, the promise of autonomic computing is obvious, but enthusiasm will inevitably be tempered by a hard-earned cynicism. "Although these efforts will be a distant reality for many organisations, the idea behind them is attractive -- that of a robust computing environment, which essentially manages itself to the level that human intervention is rarely needed," says analyst Herb Van Hook of META Group. "We believe the reality is somewhere in between the current state (intensive device-level management) and the vision by vendors promoting autonomic solutions. Self-healing and self-managing elements, components, and environments will (and are) occurring, but operational policies should still be set, and not every possible failure or performance issue can be accounted for automatically. Users should exploit the technological innovation of autonomic models where possible, but every computing environment will retain its uniqueness."

Even vendors acknowledge that businesses may be reluctant to turn over all their management functions to the IT systems that are perceived as having often let them down in the past. "It's a business issue more than a technology issue," says CA's Cooper. "There can be a reluctance to trust the system. All these systems can be set up to follow a series of actions. It's a question of whether the [human] expert wants to let that happen."

The five stages of autonomy IBM's original vision for autonomic systems identifies five basic evolutionary stages for IT systems: Level 1 (Basic) Multiple sources of system generated data, requires highly skilled staff. Level 2 (Managed) Consolidation of data and management tools, greater system awareness, improved productivity. Level 3 (Predictive) System monitors and recommends actions, staff approves and initiates actions, faster decision making. Level 4 (Adaptive) System monitors and initiates actions, staff manage against service level agreements. Level 5 (Autonomic) System monitors and initiates actions based on business processes, IT focuses on enabling business needs.

One way of developing trust in autonomic systems may be to first apply them in areas which are relatively stable, such as long-term data storage. "The majority of information that gets stored doesn't ever change," says Clive Gold ANZ marketing director for EMC. "If you want to create a system that's going to look after this information forever, you can achieve that."

However, that approach may have its limits. "One of the primary reasons for today's complexity is overspecialisation," notes IBM's Horn. "For decades, the industry has worked to solve problems at a micro level, rather than taking a holistic view. Autonomic computing is an attempt to shift gears and make sure 'smaller, faster, cheaper' is no longer pursued in isolation, but in the context of making systems work better and smarter."

Early attempts at full automation have shown promising results. Research carried out by HP Laboratories into automated storage solutions found that when fully automated, overall performance was within 15 percent of that achieved by a traditional interventionist IT manager. While the ultimate goal might be to match human performance, many businesses might accept a slight reduction in performance as a trade-off against ongoing staff costs.

Another challenge for autonomic systems is to provide integrated management over the full scope of IT activities. Analysts are cautiously confident that currently separate initiatives can eventually combine into a single system. "Linux, e-sourcing, eLiza and autonomic computing, and grid computing have incubated separately within IBM for years. [In 2002], IBM brought these initiatives into a coherent strategy," says Gartner analyst Thomas Bittman. (Incidentally, Gartner itself favours the label "policy-based computing services" rather than "autonomic computing", but that's something of a mouthful.)

IBM's original manifesto for autonomic computing defines five evolutionary stages for autonomic systems, and that vision is broadly endorsed by others in the industry. While level 5 (full autonomic systems) is the end goal, even reaching level 3 -- where recommendations are routinely made by systems monitoring software -- would represent a significant step forward for many businesses, and offer real and immediate benefits in terms of cost saving and administration.

If autonomic computing can deliver completely on its promises, then its impact could be profound, especially as businesses continue to accumulate and develop more complex IT systems. "Being able to see through the storm and quickly recognise the problem has real value," says CA's Cooper. Users working with simple autonomic systems are already reporting benefits.

As autonomic systems grow in popularity, that influence could spread. "When the systems, networks, and applications that make up the Internet don't need constant attention, the Internet can grow to serve new areas," says Concord vice president Douglas Batt. "The result is the positive influence on end users and the economy as a whole to leverage the power of the Internet for training, learning, and communicating."

Virtual Offis cuts costs with blade Managed service provider Virtual Offis has been one of the first local businesses to dip its feet into the autonomic computing pool. Indeed, the Sydney-based company is the first purchaser in the Asia Pacific region of IBM's BladeCenter server, which incorporates many of Big Blue's early autonomic technologies (we'll compare the blade server offerings from all the major vendors in next month's T&B). "While it's an emerging technology, we believe it's perfectly suited to our business model," explains CEO Craig Allen. Virtual Offis customers include Manchester Unity, Australia Post, and the Australian Institute of Company Directors. Virtual Offis runs data centres at Pyrmont and Mosman in Sydney, and Allen has been increasingly exasperated at the effort needed to maintain those servers. "As the systems get bigger and more complex, they're going to get harder to manage. It's not like looking for a needle in a haystack, it's like looking for a needle in a silo," he says. "The complexity of these systems is getting way out of hand. Most of my staff are managing basic operational issues. I would rather have those guys freed up for more important issues, and for them to be available when things are coming to a crux. I'm looking for technology where a server can fix problems by itself." With self-healing, "you free up more time for your operation," he says. "We expect to have a server setup which is like electricity -- you don't have to think about whether it's there." Allen estimates that the time taken to shift a set of applications from one server to another can be up to 2½ hours. That can prove to be an expensive prospect, especially if the cause of the failure proves to be a $10 fan. As hardware to allow failovers is already in place, the basic cost of an autonomic system is not much higher than existing systems, but should provide greater reliability. (The fan itself shouldn't prove to be a problem: the BladeCenter has two dedicated blowers, and "when things are going wrong, they sound like a Boeing taking off," Allan says.) In practice, the question of whether the server can be trusted to make the correct decisions hasn't proved to be a problem. "We're more interested in making sure that the server's available, and, if something does go wrong, it can fix itself," says Allen. "You don't want to know how it works; you just want to know that it's running." In the future, Allen is confident there'll be more sophisticated approaches to autonomic computing. "While I think the IT hardware infrastructure is moving in that direction, the operating systems and the applications are going to take a little longer. They're the next layers to get on board. The smarts are not really at the applications layer yet," although technologies such as automatic updates are an early step in that direction, he says. Ultimately, most users won't be concerned with whether a system is autonomic or not, Allen predicts. "The way that we sell it to our customers is that we're removing a layer of skill sets that doesn't offer ROI. Customers can actually get more bang for buck by using this type of technology."

Similar benefits could apply in more basic areas such as storage. "It's going to change things dramatically. People say that storage is a utility, but no other utility takes so much time to run," says EMC's Gold. "You just switch on the lights and they work, but storage isn't there yet. Businesses are not in business to run storage infrastructure, although more and more businesses exist because of information. Business wants to drive down the cost, and the whole issue with traditional storage is staffing."

"The business approach to IT will be to expect it to be a reliable tool that really helps with the aims of the business, rather than a hit-or-miss thing that causes pain, complaints and an excuse for poor performance," says Touchpaper's Ridgway.

When is it coming?

"Autonomic computing is not science fiction," proclaims IBM's Telford. "This is something that's shipping today." This is true to a degree. While there are several examples of early-stage autonomic systems, how long it will it be before such technologies are commonplace?

"You'll see examples that are very concrete start to come out first, and then you'll see these start to become more sophisticated inside the products," predicts Catherine Lord, market intelligence director for Lotus Software.

"It's moving towards having a lot more of these business rules pre-defined," notes CA's Cooper. "The goal is to provide a lot more best practices and map them back to your business practices. As people gain confidence, these systems will be applied a lot more."

One of the most fundamental moves vendors are making towards autonomic systems is better integration with existing systems management software. That approach applies even in companies like IBM which are heavily researching other areas. "We are the heart of autonomic for IBM," notes Sandra Carter, vice president for marketing at Tivoli Software, IBM's systems management division.

Vendors who aren't already on the autonomic bandwagon (under whatever name) are likely to make the shift in the near future. "Most long lifecycle, complex products like aeroplane engines and medical products are self-diagnosed and self-healing today," says Robert Lax, a partner in Accenture's communications and high-tech practice. "But manufacturers of all complex, high-tech products should be thinking about the entire service management process as customers increasingly demand lower cost of ownership around those products."

One area where most observers agree more activity is needed is in developing sophisticated self-monitoring systems within software, rather than simply basic hardware-related alerts. "This is all about software," says EMC's Gold. "You can decide to dumb everything down, so the software does the same thing no matter what is there, or you can use built-in intelligence to make sure the best facilities are used when they're there."

Improvements will also be needed in other underlying technologies. "Network performance is one of the biggest barriers there is to realising the vision of self-managing systems," says Alan Noble, chief technology officer for Foursticks. "Any autonomic computer system today must be network based. The lifeblood of an autonomic system would be the network on which it depends. End-to-end network performance management at every network node, from router to PC to server, will enable more autonomy because it enables intelligence to be delivered to the network."

META analyst Van Hook believes the uptake of autonomic functions will be gradual and predictable. "During 2002-05, IT organisations will continue to deliver basic views of abstracted data (eg, business services, business flows) to themselves and their business organisations to monitor the health and activity of their environments," he wrote in a recent analysis. "This will stay at a monitor-only level through 2005, when control functions (eg, "allocate more resources to this business service") will begin to emerge (beyond the current element level). Through 2006, task optimisation for resource change and configuration (initially, servers; later, other elements) will emerge as a top-priority effort, and logical and physical configuration databases (where the changes are initially made) will appear in 40 percent of organisations by 2007. Autonomic efforts will remain fragmented indefinitely, but will enable operational staffing ratios (eg, administrators per server) to improve 100 percent by 2008."

Like most technology visions, the most sensible approach to autonomic computing may be gentle experimentation, rather than the whole-hearted embrace of a new and as yet largely untested model. "It's a gradual road map for how you advance your IT infrastructure down this path," says Telford. "In essence, this is evolution, not revolution."

Autonomic at a glance

What is it? Autonomic computing (a term originally promoted by IBM) is based on the concept of IT systems that can intelligently monitor, manage, and heal themselves without requiring human intervention. It can be described less fancifully as self-managing systems.

How will it help my business? If the autonomic vision is fully realised, IT staff numbers and costs will be reduced, and businesses will be able to define their needs in terms of business policies rather than technological capabilities.

Do I need to worry about it now? Having autonomic capabilities in place may help immediately in cutting costs, but better implementations will become possible over the next 18 months.

What are the potential problem areas? Most current autonomic systems are fairly basic, and software integration could prove complex.

Right here, right now

Overall, the hype surrounding autonomic computing is well ahead of the reality -- but that doesn't mean we're in a total vapourware zone. Here's half-a-dozen products which claim some level of autonomic-style smarts.

Corosoft Datacenter Automation Product Suite
Linux specialist Corosoft provides a range of products designed to ensure ongoing datacentre availability without human intervention via proactive management tools. "Corporations suffer from inefficient and manual methods of administrating datacentre infrastructure," says CEO Julian Elliott.

EMC Centera
EMC's Centera storage management software monitors available storage capacity, distributes and creates extra copies of data according to demand and perceived problem areas, and ensures at least two copies of all stored data are always available. "It makes sure it's getting the optimum performance," says Clive Gold, ANZ marketing director for EMC.

Foursticks NP
South Australian network management software developer Foursticks aims to provide network connectivity solutions that can automatically vary with available bandwidth. "We don't require hard guarantees like other networking systems," says Foursticks chief technology officer Alan Noble.

HP AlphaServer
As well as being the last products actually based on the now discontinued Alpha processor line, the current line of AlphaServers from HP (via Compaq and Digital) include integrated server management technologies. "As part of our customers' adaptive infrastructure, we believe these systems are critical in meeting their needs for the next two to four years," says HP vice president for business critical systems Scott Stallard.

Lotus Domino
The first step into autonomic computing for Lotus' popular collaboration server has been in addition of Tivoli's Analyzer system to the product for load balancing. "It becomes predictive, and then it becomes self-tuning," explains Catherine Lord, market intelligence director for Lotus Software.

Touchpaper Vega
Touchpaper's IT service management product Vega includes both network management and automated help desk components. "The whole Vega suite is underpinned by the concepts of autonomic computing," says CEO Graham Ridgway.