Self-Managing Systems Not So Self-Evident

NEW YORK — Experts on autonomic computing agree: The developing sector is faced with a paucity of management standards and confusion about what vendors are currently providing to the developing sector.

During a panel discussion Tuesday about “Self-managing Systems: Views and Visions” as part of the inaugural International Conference on Autonomic Computing here, software engineers from IBM , Microsoft , HP and Amazon dispensed with potential competitive sniping in a gathering focused on their approaches to deploying autonomic systems.

The panelists were Jeffrey Frey, architect of on-demand Infrastructure with IBM; Anders Vinberg, senior architect at Microsoft; Donald Young, architect of Adaptive Enterprise; Jacob Gabrielson, principal software development engineer at Amazon; and Karsten Schwan, professor of the Georgia Institute of Technology.

Varying Approaches

Each panelist offered a window into his company’s strategies. In IBM’s case,
Frey reaffirmed that his company’s self-managing system approach is rooted
in autonomic computing, a phrase that has been used often since IBM coined
it in 2001.

Autonomic computing is a term used for a network’s ability to self-manage, self-heal and self-configure systems — be they Web services or data center infrastructure — on the fly to make sure networks suffer as little downtime
and require as little maintenance as possible. The idea is to free up
software engineers to conduct other critical tasks. Autonomic computing, Frey made clear, is part of IBM’s broad e-business on-demand computing

In this respect, HP and Microsoft concurred with IBM, but offered different approaches. For example, the HP and Microsoft representatives did not embrace the term “autonomic” but both agreed the idea of self-managing, auto-configuring systems is useful.

HP’s Young said that enterprises are faced with the task of “balancing people, processes and IT.” HP addresses this through its Adaptive Enterprise strategy, which is the company’s answer to IBM’s on-demand strategy and features automation as a key driver of self-managing systems.

Microsoft’s Vinberg agreed that automated management is important but called for a different sort of management on the strength of Microsoft’s Dynamic Systems Initiative.

While products such as HP’s OpenView and IBM’s Tivoli management software aim to explore “root cause analysis” of problems, Vinberg proposed that vendors focus on software that represses problems outright, as opposed to merely correcting them in rapid fashion. Calling for “model-based management,” Vinberg said DSI aims to provide systems provisioning, health monitoring and capacity planning to provide complete system level management.

Gabrielson from Amazon oversees thousands of servers and lines of code for the e-commerce giant. He lamented that single points of failure were problems, and said the company uses a combination of “plug-an-play” infrastructure and proprietary software to help its computers communicate.

Meanwhile, Schwan said he would like to see new computing architectures, such as advancements in Web services, grid and clustered computing, and
approved the idea of using middleware to serve as the glue between different
applications to help them speak to one another. He also said optimization
and automation isn’t as good as it should be at this point despite all the
positive talk and development.

The audience speaks

One end user asked how the vendors planned to reconcile the fact that
different systems provide data in different formats, and called for a
management standard in order to translate different formats.

Microsoft’s Vinberg replied that while one uniform standard does not exist,
there are different development tools to interconnect data, citing data
warehousing (data transformation) and the financial services industry as
examples where this is prominent. He also said management is a field so
specialized that it makes a management standard almost a moot point.

“I’ll bet most of you guys here [pointing to the panel] don’t learn from
other people,” Vinberg explained. “We have to invent everything ourselves
because management is such a specialized field.”

HP’s Young agreed. “Data has got to be collected, but I don’t see a riveting
factor as having a standard. It would make it easier, but traditional
application integration techniques used for data would apply here for the
foreseeable future.”

Frey, who repeatedly stressed IBM’s mantra of open standards, said there
needs to be a willingness in the industry for a standard to be conceived.
Chances are slim, then, if key vendors such as Microsoft and HP, don’t see
the need for one.

“If we got a little more serious about defining, adopting and implementing
standards, even if it’s through various levels of transformation and
adaptation at first, then we’d make a hell of a lot more progress than we’re
making today,” Frey said. But I also agree that there are a set of tools and
a set of capabilities that we’ve been suggestion for application
integration. We should take many of those approaches and apply them to the
management system.”

But lest anyone think the engineers are making up excuses about why there
are no management standards, Schwan came to the rescue, noting that there
are external pressures that business practices often get in the way of developing standards.

One audience member said there is a disconnect between what he heard from Microsoft’s Vinberg and IBM’s Frey on the panel.

“One is actually managing systems that are low frequency — configuration management or change [Microsoft],” he said. “The other seems to managing at
a very high frequency rate — constant load balancing, constant redeployment
of resources for optimizing business utility. These are very different types
of management.”

Frey agreed.

“If there is anything that needs to be changed in the way we deal with or
manage these systems, it’s [that] we need to embrace the notion of very loose
coupling and varying degrees of coherency across the data and the state and
the instantiation,” Frey said. “We have to build management components that
make decisions and when they execute on those decisions, have to assume that
the world can be significantly different than when they made the decisions.”

Another end user praised the vendors for their unity of vision, but asked
when we might see some products to back up the management software hype,
perhaps highlighting what could be questionable marketing between the
vendors and end users.

Vinberg noted some components, such as Systems Management Server 2003 and Microsoft Operations Manager, are
already available. HP’s Young pointed to HP’s Utility Data Center and
Virtual Server Environment, with capacity on demand exist today.

“But this is a journey, not a destination,” Young said. “We’ll continue to
strive to increase the level of automation, increase the level of adoption
and move it on up to the business level.”

IBM’s Frey pointed to several IBM products announced of late, including the
Virtualization Engine and new Tivoli provisioning enhancements. “I would
claim that IBM mainframe systems have been doing policy-driven autonomic
computing for two decades now.”

The panelists largely agreed that lifecycle
management is a unilateral goal: making sure infrastructure is cared for
from its inception in a data center until it is ready to be disposed of.

The panelists also largely agreed that it is people — not software — that are responsible for most system failures. Human error causes most of the problems they bump up against, they said. Moreover, management based on predetermined polices was also a consensus goal.

Ultimately, the panelists concluded that a lot more work is needed in
fleshing out product lines for self-managing systems and reaching out to
customers to educate them on how to run such environments.

News Around the Web