Facilitating Adaptive Placement of Management and Control Functions in Converged ICT Systems - Advances in Network-Embedded Management and Applications - page 54

a single framework. For example, managing IT and network resources
naturally differs significantly, and virtualization leads to new performance
constraints between computing and networking resources that require the
redesign of management functions.
We argue that management and control (M&C) systems must actively
support both the static nature as well as the dynamic process of convergence
in ICT systems in a way to achieve sustainable M&C for those ICT systems
that are yet to be developed. Current M&C systems, however, are not flexible
enough and competing, disparate views dominate state of the art management
and control systems.
In the quite heterogeneous landscape of M&C approaches, distribution and
hierarchy can be identified as two of the strongest architectural separators that
hinder the management and control of converged systems' performance. Let
us consider the dimension of distribution, delimited by centralized and
distributed forms of M&C. Surveying the state of the art shows that
established as well as innovative M&C control technologies prevail on both
ends. In the centralized case, classical SNMP architectures, for example, have
proven to be highly reliable, and modern control architectures, such as
OpenFlow [1], [2], show that centralized M&C continues to make sense. On
the other end, fully distributed architectures lead to M&C functions embedded
in the network elements. Two examples are the established spanning tree
protocol (STP) in Ethernet and novel in-network management (INM)
approaches [3], [4], where distributed execution of M&C functions is strongly
preferred and, optimally, does not require external intervention.
More than static choices, M&C is also characterized by transitional
developments, where, for example, handover management in wireless
telecommunication networks moves from centralized to more distributed
solutions (e.g. [5], [6], [7]). It should further be left to the system
administrator which M&C approaches he or she considers to be the most
suitable ones, depending on any of a system's characteristics (e.g. size) that
may require special consideration. Last but not least, system characteristics
will likely change over time. For example, a system that is successful will
grow in size and some of its M&C functions may have to be enhanced over
time to maintain efficient system operation.
Because fixed architectural choices put tight constraints on the distributed
and hierarchical placement of M&C functions, they provide a limit for
adaptability of converged system M&C. In contrast, there is a need for
freedom of functional allocation. Central control and management have
benefits, such as network element simplification (off-box), in terms of
information to be used for decision-making, and it is a convenient single point
for attaching policies, also high-level, such as business objectives. Strong
distribution may lead to significant gains in performance, reliability,
scalability, flexibility, and robustness [8], e.g. when functions are only locally