08-08-2012, 01:00 PM
A Cross-Layer Architecture For Autonomic Communications
A Cross-Layer.pdf (Size: 89.38 KB / Downloads: 32)
Abstract.
Layered architectures are not sufficiently flexible to cope with the
dynamics of wireless-dominated next-generation communications. Most existing
architectures and approaches depend purely on local information and provide only
poor and inaccurate information gathering at the global scale. De-layered or crosslayer
architectures may provide a better solution: cross-layering allows interactions
between two or more non-adjacent layers in the protocol stack. We propose a new
cross-layer architecture which provides a hybrid local and global view, using
gossiping to maintain consistency. We evaluate our proposal informally in terms of
communication complexity and in terms of its ability to support the “self-*”
properties being proposed within the autonomic communications community.
Introduction
The worldwide success of the Internet is dominated by the layered architecture, but a strict
layered design is insufficiently flexible to cope with the dynamics of wireless-dominated
next-generation communications. Recent studies [1] show that careful exploitation of
some cross-layer protocol interactions can lead to more efficient performance of the
transmission stack – and hence to better application layer performances – in different
wireless networking scenarios.
Cross-layer design breaks away from traditional network design, where each layer of
the protocol stack operates independently and exchanges information with adjacent layers
only through a narrow interface. In the cross-layer approach information is exchanged between non-adjacent layers of the protocol stack, and end-to-end performance is
optimized by adapting each layer against this information Cross-layering is not the simple
replacement of a layered architecture, nor is it the simple combination of layered
functionality: instead it breaks the boundaries between information abstractions to
improve end-to-end transportation.
Related Study
Research on cross-layer networking is still at a very early stage, and no consensus exists
on a generic cross layer infrastructure or architecture. However, the importance of a sound
architecture to handle the proliferation of cross-layer operations in wireless as well other
communications media is clear, especially in autonomic systems for which properties
need to be specified and maintained with minimal manual configuration and intervention
[11]. A number of proposals for cross layer designs and their corresponding architectures
have been published in the literature. Most of these architectures are relying on node-local
view and very few utilize both the local view as well as network-wide global view.
GRACE (Global Resource Adaptation through Co-opEration) [6] is a cross layer
adaptation framework. All system components (hardware, network, and operating system)
and applications are allowed to be adaptive. These adaptive entities co-operate with each
other to achieve a system-wide optimal configuration, for example to maximize system
utility, in the presence of changes in the available resources or application demands.
However, its cross-layer approach includes no explicit consideration of cross layering
within the networking layers or protocol stack. WIDENS (Wireless DEployable Network
System) [7] has been proposed with an aim to acquire the interoperability, cross layering
and re-configurability at the same time. This cross layering architecture seems a
promising one where protocol optimization is based on the local state information but it is
still in the validation stage and so lacks any real measurement of efficiency especially in
terms of performance.
Cross-Layer Architecture Based On a Global View
Both OSI and TCP/IP adopt a bottom-up approach driven by physical and network
constraints, which can make it hard to capture and respond to top-down user demands or
contexts. Cross-layer design can help capture these concerns by providing a more uniform
framework within which to capture and disseminate information at different semantic
levels. Realizing the importance of cross-layering – and specifically cross-layering
architectures with a network-wide global view – in next-generation communications, we
propose an alternative cross-layering architecture based on a combination of node-local
and network-global views. “Global view” is a broad term suggestive of centralization and
reduced scalability, but we will try to establish and maintain a network-wide view of
multiple metrics depending on the focus of the network without undue impact. Metrics
such as load, battery status and so on in can be very useful in attaining self-organization or
self-healing in an autonomic network, and having a global view allows a node to evaluate
its own status against the average within the network at any instant. For example, a node
could decide whether it is carrying more traffic then the average node and how overloaded
it is compared to the average, and could utilize this information for routing and load
balancing.
Overview of the Architecture
Alongside the existing layers, the Knowledge Plane is the key element of the architecture.
Direct communication between layers and a shared knowledge plane across the layers are
the two widely used cross-layer interactions polices. Because of the improved separation
and management possibilities we prefer to utilize the knowledge plane for the
architecture. The following are the main elements of the architecture:
Existing TCP/IP layers: These provide normal layering support when it is necessary,
as well the information to the knowledge plane related to different layers to maintain local
view of the node. This allows full compatibility with standards and maintains the benefits
of a modular architecture, as it does not modify each layer’s core functionality.
Contextors for different layers: Each layer in the existing protocol stack has a
corresponding contextor, which will act as their corresponding interface between the layer
and the knowledge plane. Each of these contextors acts as a “tuner” between a layer and
the knowledge plane. Possible functionality for manipulating protocol data structures is
built into the contextors: no modification is required to the existing protocol stack. This
facilitates incorporation of new cross layer feedback algorithms with minimum intrusion.
Generally a protocol implementation has data structures for control and data, with a
protocol’s behavior being determined by its control data structure. A contextor is
responsible for reading and updating the control data structures when it is necessary.
Conclusions and Future Work
The worldwide success of the Internet has led to the domination of the layered
architecture, but a strict layered design is not flexible enough to cope with the dynamics of
next-generation communications. Moreover exploitation of some cross-layer protocol
interaction can lead to more efficient performance of the transmission stack (and hence
better application layer performances) in different wireless networking scenarios. Most
existing architectures depend on the local information and only CrossTalk depends on
local as well as network wide global view. Even with the uncertain and poor global view
gathering process, CrossTalk has shown performance improvement in load balancing
algorithm. With a more accurate global view gathering process we can hope for further
improvements, and with this in mind we proposed an alternative cross-layering
architecture for autonomic communications which is based on local information as well as
global information and gossiping will be the mechanism to collect the global view related
information. The use of gossiping provides more impact that piggy-backing but keeps
overheads more controlled than flooding, and allows the time constants and latencies of
information to be varied widely.