NMRG/EMANICS Workshop Position Statement

Olivier Festor <olivier.festor@loria.fr>
LORIA - INRIA Lorraine


Autonomous/mic Management

Automating as much as possible the management functions is not new in
the device, network and service management community. In fact,
automation is a goal behind almost every evolution in the control and
management plane. A good illustration of this is, as presented in the
DSOM'2003 panel on self-management by Alexander Clemm, the evolution
of address configuration procedures which evolved from manual
configuration, to fully distributed stateless configuration, through
semi-automated allocation with DHCP. Many other such examples of
evolutions exist in the Internet.

Continuing to improve automation in the management plane is and will
remain a major challenge in the next decade. Open challenges in this
area are: (1) ensuring traceability of the actions taken by the
autonomous management plane, (2) ensure safety of the autonomous
management plane (provide guarantees, even statistical ones, that the
system brings/maintains the managed environment in a stable and better
state after management actions have been taken).


Scale

The number of communicating devices, the networks that interconnect
them and the services that run on top of them are increasing at a
very fast rate often in constrained environments. This growth in
size, often combined with increasing constraints on the underlying
infrastructure (power, connectivity, CPU limitations) puts strong
requirements on the management plane. While some performance
evaluations were reported in a couple of contributions published in
the last decade, better understanding, modelling and evaluation of
scale in the management plane is still necessary.       

Recent work on this topic is very promising (management trace
analysis, behavior modelling, and benchmarking activities in the
EMANICS NoE, large scale systems update analysis by Gkantsidis & al.
in SIGCOMM 2006). Such efforts need to be continued and extended.
Some open challenges in this area are : (1) the definition of common
metrics and benchmarks like they exist in the database community, (2)
the design and operation of large testbeds to enable reproducible
experiments, (3) in depth study of large scale configuration as well
as fault management activities.


Managing in Hostile Environments

Networks are getting more and more hostile to traditional management
approaches. Entities like firewalls, NAT devices, problems like
security flaws in some management approaches or the way they are used,
dynamics of managed devices, changing connectivity conditions or
simply the absence of support (agent) for a standard management
approach make the life of traditional management very hard.

Management data is still available in the devices, networks and
services, but the way it is accessed, processed and used is currently
changing drastically.  It is a fact that management functions get more
and more embedded and take various new forms (trust, reputation
schemes, incentives, ?) . This mutation of management functions is
very promising. It also represents a real challenge for the next
decade.  Open challenges in this area are: (1) design of new
?opportunistic? management models and algorithms for specific
management functions on specific problems that use alternative ways to
collect/process management data and act on the managed environment,
(2) increase investigations on ?probabilistic management?, well
adapted to networks which exhibit strong dynamic behaviours.