Autonomic communication QoS - fixed and/or next generation networks Kaisa Kettunen

advertisement
Autonomic communication QoS
- fixed and/or next generation networks
Kaisa Kettunen
Helsinki University of Technology / S-38.4030
Seminar 26.-29.5.2006
What is Autonomic Communication (AC)?
Current trend in networking:





High amount of new services and applications
Increasing network sizes with increased management costs
Deployment and introduction of new networks (e.g. Next Generation Networks (NGN))
High complexity of network infrastructures
Growing demands (e.g. security, quality, resilience, reliability etc.)
There is a pressure towards converged network architecture (heterogeneity) with
open interfaces and standard protocols.
Autonomic communication – the vision:
”Towards self-organizing, self-managing and context-aware autonomous network”
By moving intelligence to the network, create an adaptive and aware environment to
achieve a common purpose by self-organization of network nodes and consequently
reduce network management complexity and human intervention.
Kaisa Kettunen
Helsinki University of Technology/S-38.4030
2
AC vs Quality of Service (QoS)?
Quality of Service (QoS) is the level of performance that is expected by a user in a
network. It is defined by the network performance as well as performance of the used
service components and can be characterized by for example the following:





Bandwidth (bps)
Delay (ms)
Loss rate (%)
Jitter
Noise/Loudness level
End-to-end (e2) QoS is the overall performance of all involved networks and service
components.
Assuring wanted e2e QoS with ever increasing network interworking requires dynamic
QoS management, which makes use of the existing network synergies and distributes the
needed tasks in order to reach the wanted outcome.
 Autonomic Communication enables this!
Kaisa Kettunen
Helsinki University of Technology/S-38.4030
3
Example of Autonomic (QoS) Provisioning
Framework
InterDom
Composition
Adaptation
User
Knowledge
Domain
Discovery
IntraDom
Inter-Domain
Monitors
Cross-DomainCross-Domain
Provisioning Contracting
IntraDom
IntraDom
Intra-Domain
Provisioning
Intra-Domain
Provisioning
Intra-Domain
Provisioning
Knowledge
Knowledge
Knowledge
Intra-Domain
Monitors
Resource
Allocation
Intra-Domain
Monitors
Resource
Allocation
Resource
Allocation
Intra-Domain
Monitors
┤├
┤├
┤├
QoS Monitor
QoS Monitor
QoS Monitor
Source: ”QoS-Aware service composition and adaptation in Autonomic Communication”, J. Xiao & R. Boutaba
Kaisa Kettunen
Helsinki University of Technology/S-38.4030
4
Autonomic component key functions
Each domain (IntraDom) has an own QoS provision mechanism and offers a set of service
classes.
The three functions of an autonomic component are applied to the inter-domain component
(InterDom):
Sensor
Domain
Discovery
Inter-Domain
Monitors
Used to observe and to report data about aspects of the system, e.g.
reachability of nodes or monitoring aggregate QoS service condition
Actuator
Cross-DomainCross-Domain
Provisioning Contracting
Used to change behaviour, e.g. establishment of QoS contracts with each
domain or obtaining a QoS-assured path segment at specific border points
Analyzer/planner
Composition
Adaptation
Kaisa Kettunen
Used to ensure required e2e QoS based on cognitive computations
Composition: What networks or service providers to involve?
Adaptation: Dynamic adjustment of network/service composition based on
communication path monitoring and re-evaluation to ensure wanted outcome
Helsinki University of Technology/S-38.4030
5
Requirements for adaptation and composition
algorithms
The performance of an adaptation and composition algorithm complies ideally to
the following requirements:
1.
2.
3.
4.
Reasonable processing time enabling fast response
Minimized cost and degree of disruption during processing
High probability of finding a feasible path with near optimal costs
No excessive communication overhead introduced
Note that the efficiency of the algorithm also depends on the Sensor and Actuator
functions.
On the following slides, examples of published methods to improve Quality of
Service with the principles of Autonomic Communication in different kind of
networks are presented.
Kaisa Kettunen
Helsinki University of Technology/S-38.4030
6
Cognitive Packet Network (CPN) routing
CPN is a distributed protocol for packet networks, which provides dynamic routing based
on sensing and monitoring and which is driven by a QoS goal defined by the user or by
the network itself.
Smart / Cognitive Packets (SP) are used to discover routes and collect measurements (no
payload).
Acknowledgement Packets (ACK) are generated at destination as response to received SPs
to carry back the original packet route and the measurement data along a reverse route
established by removing any sequences which begin and end in the same node. Example:
Original route:
Reversed route:
<a, b, c, d, a, f, g, h, c, l, m>
<m, l, c, b, a>
Dumb Packets (DP) carry payload and use the source routing.
Mailboxes (MB) in nodes store the QoS information carried by ACKs per QoS class and
destination. The information is organized as a Least-Recently-Used (LRU) stack (new info on
top).
Reinforcement Learning (RL) is a Random Neural Network based algorithm running at each
router. It uses the observed outcome (SP success/failure) of a previous decision to ”reward”
or ”punish” (weight factor increase/decrease) the previous (link=neuron) choice based on the
set goal.
Kaisa Kettunen
Helsinki University of Technology/S-38.4030
7
Experiments on CPN routing
Several QoS experiments with a number of nodes have been published with a setup similar to
the one below. Here, each pair of nodes is connected by P2P 10Mbit Ethernet links. The
system is fed with UDP packets with a constant bit rate and inserted random background traffic
on the links.
CPN Node
201
Source
CPN Node
202
CPN Node
203
CPN Node
204
CPN Node
205
CPN Node
206
CPN Node
208
CPN Node
209
CPN Node
210
CPN Node
211
CPN Node
212
CPN Node
214
CPN Node
215
CPN Node
216
CPN Node
217
CPN Node
218
CPN Node
219
Destination
Conclusions:

CPN can approximately find shortest path as well as offer more complex QoS criteria (e.g.
delay) for routing. Using more complex criteria than the shortest number of hops can provide
better overall quality of service.

A comparetively small fraction of SPs and ACKs of total user traffic is needed to serve a user
Goal and a small number of SPs can suffice to initially set up paths.

Usage of Genetic Algorithms for finding new routes improves QoS under light network traffic
but not under high traffic conditions (where it seems to slow down the adaptation logic).
Kaisa Kettunen
Helsinki University of Technology/S-38.4030
8
Autonomic distribution in Peer-to-peer (P2P)
networks

Peer-to-Peer networking is based on collaboration, intercommunication and resource
exchanging among individual nodes. It provides a highly dynamic environment with
unpredictable quality of service.

In absence of a centralized resource management, autonomic distribution based on
regulation and rules can provide better QoS by self-organization of a peer community
in terms of e.g.



availability, capacity and memory
load balancing
reduction of redundant data storing
”Contribute while consuming”

Example: SelfService


Protocol based on trial&error, local memory and broadcast requests for sharing application
modules between individual machines based on simple reasoning of every peer.
This again requires improved information collection mechanism

Two common ways: explicit probing, e.g. hearbeat messages (-) or multicasting (+)
Kaisa Kettunen
Helsinki University of Technology/S-38.4030
9
New protocols for improved P2P QoS

PeerWindow is an information collection protocol, with which each node can collect a
large amount (N/2level) of information pointers to other nodes at a low cost.



Self-determined level of a node defines its capacity in relation to other nodes and the size of
the node’s peer list containing pointers to other nodes. In an k-level node, the list should
contain pointers to all nodes whose nodeId’s first k-bits (eigenstring) are the same with local
one.
All the nodes, whose peer list contains a pointer to a given node, form a set - node’s audience
set, which must be informed at changes.
Autonomic group communication protocol can be used to
guarantee a QoS required by applications also if the QoS
supported by the underlying network changes

Protocol modules are realized as autonomous agents, which
change classes, i.e. ways to implement a protocol function,
based on monitored QoS
Kaisa Kettunen
Helsinki University of Technology/S-38.4030
10
QoS improvements for Wireless networks
In wireless networks performance is measured on e.g. power, range and data range. The
“chaotic” deployment may be improved with Autonomic Communication in terms of

Automated power control and rate adaptation to minimize interference between neighbouring
access points (AP) by reducing power to the minimum level which allows reaching receiver at
maximum transmission rate

Load management and effective coverage

Adaptive traffic scheduling mechanisms used in case of network changes or according to
application needs to save node energy and to avoid overload as well as compulsive behavior

Nodes classified to zones and further grouped as virtual sectors. An intelligent migration agent
monitoring the network activity can order a change between interchangeable scheduling tables (e.g. Xor V-scheduling) for an entire sector or zones in a sector.
For ad hoc networks formed by passerby peers with no central control, Peer-To-Peer
Wireless Network Confederation (P2PWNC) has been developed

Scheme where a set of administrative domains provide wireless service, e.g. Internet access, to each
other’s broadband Wi-Fi users. This is done based on an algorithm that detects non-simultaneous
multi-way P2P exchanges.
Kaisa Kettunen
Helsinki University of Technology/S-38.4030
11
QoS in Next Generation Networks (NGN)

For next generation networks, a more broader view beyond a specific environment is
needed to establish a working reliable co-operation.

The concept of Ambient Networks (AN) aims to improve interworking of different
environments for e.g. self-organized establishment of QoS


ANs agree to follow composition agreements, e.g. to maintain QoS for a mobile video
conference in a moving train, with means of topic-related control and management tasks
provided by a Functional Area (FA). Control functions may be distributed and a procedure is
independent from the nature of the entities involved.
In a SIP-based network, autonomic communication can be used for fault
recovery/avoidance as well as dynamic load balancing

Through monitoring events over Service Bus, node loads can be supervized and a ”Recovery
Agent” logic may allocate or restart services from suitable servers.
Kaisa Kettunen
Helsinki University of Technology/S-38.4030
12
Conclusions

Controlling end-2-end QoS in current static and manually configured
network environments is becoming increasingly complex

Principles of autonomic communication provide means to improve
QoS in existing networks and enable maintaining it in the next
generation solutions

There is a variety of dynamic QoS related protocols, algorithms and
schemes which have been published and evaluated successfully so
far and more is yet to come…
Kaisa Kettunen
Helsinki University of Technology/S-38.4030
13
References

QoS-Aware service composition and adaptation in Autonomic Communication , J.Xiao and
R.Boutaba, IEEE JSAC, Vol. 23, No. 12, Dec 2005


QoS and Routing in the Cognitive Packet Network, E.Gelenbe and P.Liu, IEEE, WoWMoM’05
Autonomous Smart Routing for Network QoS, E.Gelenbe, M.Gellman, R.Lent, P.Liu and P.Su,
IEEE, ICAC’04

Self-Aware Networks and QoS, E.Gelenbe, R.Lent and A.Nunez, Proceedings of the IEEE, Vol. 92,
No. 9, Sep 2004


”Self-Service” – A Theoretical Protocol for Autonomic Distribution of Services in P2P
Communities, F. Saffre and H. R. Blok, IEEE, ICAC ’04
PeerWindow: An Efficient, Heterogeneous, and Autonomic Node Collection Protocol , J.Hu,
M.Li, H.Dong and W.Zheng, IEEE, ICPP ’5


An Autonomic Group Communication, T.Enokido and M.Takizawa
Self-Management in Chaotic Wireless Deployments , A.Akella, G.Judd, S.Seshan and
P.Steenkiste, MobiCom ’05

A Self-Managed Scheme for Free Citywide Wi-Fi, E.C.Efstathiou and G.C.Polyzos, IEEE,
WoWMoM’05

Adaptive Scheduling in Wireless Sensor Networks, A.G.Ruzzelli, M.J.O’Grady, G.M.P O’Hare and
R.Tyran, Department of Computer Science, University College Dublin

A Framework for Self-organized Network Composition, C.Kappler, P.Mendes, C.Prehofer,
P.Pöyhönen and D.Zhou,

Towards service awareness and autonomic features in a SIP-enabled network, G.Delaire,
L.W.Goix and G.Valetto, Telecom Italia lab
Kaisa Kettunen
Helsinki University of Technology/S-38.4030
14
Download