Quality of Service in IN-home digital networks Alina

Quality of Service in IN-home digital networks Alina Albu 23 October 2003

Contents Presentation: “Prediction-based policy adaptation for Qo. S management in wireless networks” Proceedings of the 4 th International Workshop on Policies for Distributed Systems and Networks (POLICY ‘ 03)

Current situation, demands n mobile end users having Anywhere/ Anytime access to a wide range of computing services, with an emphasis on multimedia applications (video tele-conferencing and news-ondemand)

Current situation, challenges Delivering seamless Qo. S based on individual users needs (the network must provide different levels of service to different categories of customers).

Challenges Caused by: n n n wireless network component characterized by lower bandwidth and greater packet loss rate. The quality of service of the wireless component may change abruptly over time due to geographic impairments, weather conditions. Mobile users can move between cells characterized by different number of served users and hence, with different available bandwidth.

Challenges n n Even if users remain static the bandwidth available to them may vary due to the mobility of the other users. Due to users’ movements, the path between the sender and the receiver may change => cause a rerouting of data packets => a possible variation of resources even in the wired part of the connection.

The current solution n The selection of a strategy for Qo. S management is usually statically fixed

The need for a new solution n Dynamic Qo. S management approaches are needed that can support the dynamic variation of network strategies.

New proposed solution n Framework for policy-based management that defines a set of components to enable policy rules definition, saving and enforcing.

Policies n Set of predefined rules that govern network resources, including conditions and actions with parameters that determine when the policies are to be implemented in the network

Proposed Framework n n Multi-Agent System in which intelligent agents cooperate to predict future changes in the delivered Qo. S and adapt the network behavior according to these changes. Architecture composed of 3 layers of agents: n n n Qo. S prediction agents (QPA) Qo. S adaptation agents (QAA) Monitoring agents (MA)

Proposed Framework

Qo. S Prediction Agents (QPA) Utilize user’s information such as: n User profile n Location n Terminal characteristics to predict possible future changes that would affect the delivery of the required Qo. S level.

Qo. S Prediction Agents (QPA) Prediction performed by 3 different agents: n n n User Agent (UA) Location Agent (LA) Application Agent (AA)

User Agents (UA) n User’s actions and preferences play a role in the process of future predictions. Users can specify their Qo. S requirements through the use of User Policies (UP). n n The UA can access the UP along with information about the user’s preferences and terminal characteristics. UA is responsible for analyzing this info and predicting any events that would trigger a UP. Policy - triggered, -> UA responsible for reporting the necessary actions to be taken by the QPA.

Location Agent (LA) n The LA is responsible for predicting the future location of the user, assessing its effects on the Qo. S, and delivering the info to the QPA.

Application Agent (AA) n Similar to the UA, the AA is responsible for analyzing and specifying the continuously changing Qo. S requirements for each running application.

Qo. S Adaptation Agent (QAA) n n n The goal of the QAA is to provide the required Qo. S based on specifications from the QPAs, taking into consideration current and predicted users/network statuses. The specifications are given in the form of sets of network level adapted policies to be applied to the network resources. The QAA assembles sets of policies at run-time, dependent on the sets of imposed constraints and goals that need to be satisfied.

Monitoring Agents (MA) n n MA provides a real-time feed-back for the QAA. It is responsible for the measurement of the Qo. S exhibited by the network. An MA may interact with other Mas in order to gather the required measurements of Qo. S. Obtained measurements are then reported back to the QAA. The QAA is responsible for instructing the MA with the necessary parameters to be measured.

Policy Adaptation n Types of policy adaptation: n n n Adaptation carried out by dynamically changing the parameters of a Qo. S policy to specify new attribute values enabling/disabling a policy from a set of predefined Qo. S policies at run-time. Learn from the current system behavior and create new policies at run-time

Policy Adaptation n n Policy adaptation performes by QAA -> belongs into the 3 rd category QAA architecture, scenario: n Stage setting n Candidate actions selection n Policy assembly n Reassessment

Stage setting n n n QAA specifies its objectives, constraints, actions set, definition of success The objectives - determined based on info obtained from the QPA representing values for the required Qo. S parameters. The constraints set represents constraints imposed by the device features (memory size, CPU speed, available software). Actions set – possible actions that can be used in the policies’ action part Each action – associated an utility function ->expresses the degree of usefulness of the action. Utility function ->function of Qo. S parameters (delay, jitter, throughput)

Candidate Actions selection n QAA selects one or more actions set which would best attain the specified objectives. QAA may negotiate with other neighboring QAAs for the selection of the most suitable actions to be taken Through the negotiation, actions with the highest utility values are selected.

Policy assembly n n Assembly for one or more network policies given the actions selected in the previous step. Each assembled policy = triggering event, a set of conditions, actions Each policy – associated with a life time after which it should expire and be deleted. Once a policy is assembled it is dispatched to be applied at the network level.

Reassessment n n Evaluate the degree of success of the previously dispatched policies. Based on the comparison between the Qo. S measurements (provided by MA), and the current definition of success, the reassessment module – decide to modify the utility function associated with the actions already applied -> either increase or decrease their value.