broadband and ip metrics: modeling, measurement and optimization


Overview - Description




Telcos/network operators/ISPs offer data and voice services to millions of customers. Network operators and service providers aim at optimizing the process of continually upgrading their network infrastructure to support new services and/or new customers. Adding new network components or new services is a significant investment. To predict/analyze future profit from an investment (i.e., whether adding new services to the networks would be economically viable) is a complex process. Without sound methodologies to characterize network performance/reliability, without accurate data about service availability/quality and without the tools to collect and analyze these data effectively, network planners will find it increasingly difficult to make intelligent quantitative network decisions.

The goal of this project is twofold. First, to define effective metrics that could help shed the light on the health of the network and of the services provided to customers. An important aspect here is to determine the dependencies/correlation between technology, network, service and customer levels metrics. Secondly, to develop the tools to measure, correlate and report these metrics in a multi-services network composed of heterogeneous devices. The data will then be analyzed and used in decision-making processes. The leading focus of the proposed research is to enable automation of capacity planning and network optimization.

Issues to be studied in this project include:

  • Empirical versus formal metrics

  • Composition of metrics

  • Errors, uncertainties and sanity checks

  • Measurement strategies and infrastructure

  • Methodology (process of quantifying a measurement for a metric)

  • Use of standard metrics (e.g., IETF IP performance metrics).

  • Frequency of measurements.

  • Data collection and management (placement of monitors, storage, access, etc.).

  • Real-time issues.



The Internet has become increasingly vital during the last years. The next generation Internet is expected to provide service differentiation and support emerging quality-of-service sensitive applications such as VoIP and VPNs. However, the Internet has grown so complex that even well-informed users and engineers have only a fuzzy understanding of its topology, of the paths taken by their data from one Internet site to another, and of the reliability and performance of those paths.

Without sound methodologies to characterize network performance/reliability, without accurate data about service availability/quality and without the tools to collect and analyze these data effectively, network planners will find it increasingly difficult to make intelligent quantitative network decisions.

The project is structured as follows:

Technology-level Metrics:

Technology management layer deals with monitoring and controlling individual network elements such as routers, switches, physical links, etc. In this project emphasis lies on technology-dependent performance monitoring through existing MIBs and probes as well as the standard quantification of measurements for technology-level metrics. Examples of technology-level metrics include:

  • Link Metrics (propagation time, transmission time, effective bandwidth, etc.)
  • Router/switch Metrics (buffer size, queue size of a router interface, congestion-level, route processing delay, etc.)

Network-level Metrics:

Network management layer is responsible for ensuring that the network infrastructure supports the end-to-end delivery of the required services. This research focus is on the data collection, analysis and composition of network level metrics. This should consider a layered network architecture including: Optical paths, ATM/FR links and IP routing. An example classification of network-level metrics is:

  • Routing Metrics (announced routes, route flaps, stability, reachable destinations, etc.)
  • Path Metrics (delay, flow capacity, mean packet loss, mean RTT, jitter, hop counts/congestion, etc.).
  • Security Metrics (user identification/authentication and access control, firewall protection, data integrity, DNS integrity, routing table integrity, router/switch security, auditing, etc.).
  • Other Metrics (access capacity, connect time, total traffic, peak travel, etc).

Service-level Metrics:

Service management layer focuses on service installation, engineering and delivery. In this project emphasis lies on the different metrics that a Service Level Agreement (SLA) may specify. Examples include:

  • Availability Metrics (Number of users that can be served simultaneously, Dial-in access availability, Outage duration, Mean time between failures, Mean time to restore, Error rate, etc.)
  • Responsiveness Metrics (Response time, Average one-way delay/latency, Average round time delay/latency, Help desk response time, etc.)

Customer-level Metrics:

Customer management layer involves direct interaction with end customers to provide, maintain, report on service, and bill for network services. In this project emphasis lies on the different metrics involved in a customer care process. Examples include:

  • Service Fulfillment Metrics (timely delivery of what the customer ordered, etc.).
  • Service Assurance Metrics (maintaining the service--timely response and resolution of customer or network triggered problems, managing and reporting performance for all aspects of a service, etc.)
  • Billing Metrics (timely and accurate bills, invoicing, timely adjustment handling and payment collections, etc.)

Linkages between these Metrics:

In order to establish the metrics tree and thereby facilitate the planning and design process and enable its automation, it is important to set up the linkages between the metrics used at the technology, network, service and customer layers. Such linkages will allow aligning and filtering lower layer metrics with those higher in the process as well as providing feedback on the usefulness and completeness of the metrics. This activity involves the definition of interfaces between the tree levels for metrics composition as well as the use of customer, network, service and technology metrics by multiple network planning and design processes.

Figure 1 illustrates how a customer level metric can be related to service, network and technology level metrics. This gives an idea of how problems can be identified inside a network that hampers customer perception of the service.



Figure 1: An illustration of how metric relations can help in problem identification


Letís consider the following scenario: customer A tries to access his/her email account hosted on a ISP email server, but is unable to retrieve any email. Customer A phones help desk and complains about this incident. The metrics model can quickly isolate the cause of Customer Aís problem to be delay related and further track the problem down to a particular link in the network. Without the metric tree model, such problem diagnostic process could take a team of network experts hours to isolate and identify.