Filter
Top Products
Converged network design
Issue 6 January 2008 305
train. A conscious choice to favor simplicity in design may be the single biggest factor in
improving uptime due to its cascading effect on process, documentation and verification.
Design for Manageability
Studies of operator errors have identified several classes of errors typical of network service
administrators. Most of these are the result of misconfiguration of new components and
unintended actions such as restarts or disabling of hardware while diagnosing problems.
Significantly, operators of all experience levels were found to introduce almost all classes of
errors with roughly equal frequency.
Research conducted at Rutgers University [6] found that operator action-verification techniques
allowed detection and prevention of over half of the errors typically introduced by operators.
This data argues strongly for the implementation of reliable change control procedures, and
change verification as requisites for highly available networks. To support these activities
management tools must be in place to aid in detecting and reporting errors, both for validation
of operator actions and diagnosing problems. Network documentation is typically inaccurate
and outdated [7] (due in part to lack of change control) so management capabilities to verify
configurations are essential.
Design for Scalability
Other researchers have proposed mechanisms for reducing or eliminating the need for operator
interaction by automating common tasks. The success of these approaches argues that
reducing the scope of changes required to manage and expand network services will pay
dividends in network uptime. Excess bandwidth, unused ports and available addresses are
required to verify changes and to simplify network expansion. Expansion should begin well
before these resources are exhausted.
Using designs that limit the impact of changes reduces the potential for errors. For example, if
the administrator needs to change both an aggregation switch and a router configuration in
order to accommodate new media gateway interfaces, the design has doubled the opportunity
for configuration error. If a new subnet needs to be added to Access Control Lists throughout
the network core the potential for outage is expanded further still.
The same principles used to reduce software complexity and improve software reliability are
applicable to the network complexity problem. Modularity, design reuse, and testability are all
attributes of highly reliable networks.