Facilitating Adaptive Placement of Management and Control Functions in Converged ICT Systems - Advances in Network-Embedded Management and Applications - page 63

„traditional“ control applications „traditional“ management applications

Applications

„traditional“ management paths

„traditional“
protocol
stack

TCP/IP
network
interface

management
node interacts
in a traditional way with other
DC nodes and also with self-
organizing f unctions

controller activates
and interacts with
self -organizing control /
management f unctions

management
domain

OFC

control domain

NW domain

VMs

VMM
switching

switching

network
interface

network
interface

IT domain

network switch (OFS) physical server

implant leightweight, task-specific and self -
organizing control / management f unctions

OpenFlow switching on physical servers
would for NW/IT joint self -organization

task specific and self -organizing control /
management f unctions on application layer

Fig. 4. Application of management capabilities to a converged IT/network system.

In this figure, SNMP in a weakly distributed form, OpenFlow centralized
control, and INM distributed network management with objectives on the IT
side are all combined in a single architecture. The figure shows in particular
how centralized approaches merge with decentralized management functions,
that is, INM. To provide uniform handling of the technologies, adapters are
used that encapsulate e.g. SNMP agents to allow standardized communication
between management capabilities of other types of management and control
protocols. Note in particular the resulting homogeneous switching layer,
where each network element (including servers) supports OpenFlow
switching functionality homogeneously.

4.2 Scenario Description

The scenario we consider is a data center with IT and network resources,
and virtualization. We consider two use cases that are closely linked: 1)
anomaly (congestion) detection, and 2) virtual machine migration with flow

rerouting that follows after an anomaly has been detected (also see Fig. 5 ).

Assumptions

1) An anomaly occurs locally, e.g. due to the exhaustion of CPU capacity

at a host because of high load that is incurred by virtual machines.

2) Local performance checks find out about other physical hosts to which

some virtual machines may be migrated.