Facilitating Adaptive Placement of Management and Control Functions in Converged ICT Systems - Advances in Network-Embedded Management and Applications - page 64

3) One or more virtual machines are migrated, and flows are automatically

adapted without contacting the central OpenFlow controller. This
requires that a subset of controller functions must be implemented in
the host initially, that is, the hypervisor that is executed on the physical
machine.

4) In a transitional step, the required OpenFlow controller functions are

moved from hypervisor space to the programmable network interface
for improved efficiency. This step is the offloading of network
functions that follows the idea of [32].

optional coordination

management and
control domain

management nodes
(e.g. OF controller)

only relevant changes
are signalled to
management system
(e.g. new VM locations
to load balancer)

only high-level
changes are signalled
to OpenFlow
controller (e.g.
f reed switches)

VM migration triggers
f low rerouting only
in relevant
network switches

f lows are rerouted
transparently by
distributed coordination
between switches

self -organization and
„tranditional“ control /
management can
be mixed arbitrarily

S 1 S 2

detection of VM
migration possibility
to other segment
using coordinated
calculations (e.g.
based on VM load)

flow
reconfiguration

network
switch

S 3 S 4 S 5 S 6

S 7 S 8 S 9 S 10 S 11 S 12 S 13 S 14

VM coordination f or
a single type of server
application handled
only between relevant
physical machines

VM migration

H 1 H 2 H 3 H 4 H 5 H 6 H 7 H 8

Fig. 5. Converged IT/network scenario in a typical data center .

In the described scenario, it is essential that some modifications in the
deployment, e.g. from host (hypervisor) space to NIC space, are likely not to
be available immediately but require upgrading on the hardware side, for
example. A typical occurrence of events is that during the detection of
anomalies at a physical host, the management system that is currently
deployed in the data center sets a limit on the maximum number of virtual
machines that may be deployed on a host. With the availability of more
advanced NICs supporting programming, the transitional step in 4) can be
applied after the redeployment of the corresponding management capability to
the programmable NIC execution environment.

4.3 Discussion of Scenario

In order to evaluate the feasibility and benefits of proposed framework for
adapting the placement of management and control functions, we have
implemented a simulation prototype for simulating large-scale data centers,