Facilitating Adaptive Placement of Management and Control Functions in Converged ICT Systems - Advances in Network-Embedded Management and Applications - page 65

and specifically, data center networks. The prototype is able to perform
migration of virtual machines and flows by cooperation between network and
virtual machines.
In previous work, the advantage of different architectural choices for
network management and control in terms of hierarchies and distribution has
been shown to a large extent in quantitative terms. For example, Du et al. [28]
evaluate the benefits of a distributed agent-based management approach
versus a centralized approach, where the distributed architecture incurs
significantly smaller communication cost in terms of number and size of data
packets compared to a centralized solution. Furthermore, most of the
architectural choices that the discussed frameworks and algorithms in Sec. 2
follow are motivated by certain desired performance characteristics.
Rather than providing additional quantitative results, we focus here on
evaluating the central claim of our framework, which is the adaptive support
for functionality placement along the hierarchy and distribution dimensions.
From the scenario that we have implemented in the simulator prototype, we
can summarize the following points where dynamic placement is highly
beneficial. We focus on a number of concrete practical situations of the
described scenario:

• Pushing OpenFlow control functionality down the hierarchy: in the

first OpenFlow implementations, network control remains within the
OpenFlow controller. The scenario has shown that dynamic placement
is highly desirable to push OpenFlow controller functions down to
network elements. Moreover, it is beneficial to do this in at least two
steps. In the first step, control functions to manipulate network flows
can be pushed to a host (e.g. to hypervisor space). In a second step,
control functions can be pushed further to the programmable NICs. Our
framework supports this function pushdown on the level of individual
management capabilities.

• Increasing scalability with scenario size: When the size of the

scenario changes, e.g. when additional network elements and physical
servers are added, it might become necessary to change function
placement in order to maintain scalability and to avoid bottlenecks
incurred by management and control functions. It is easy to expand the
management and control structures via our framework in this case by
extending the structures on the collaboration and/or organization
direction by introducing additional management capabilities. Assuming
deployment frameworks are used that support discovery and other basic
runtime functions, which are provided e.g. by OSGi, it is
straightforward to adapt structures dynamically using the mechanisms
provided by the proposed management capabilities.