Flow Analysis in Network Analysis and Design Guide

Components of the Network Analysis and Design (NAD) Process

The Network Analysis and Design process follows a structured systems approach. Flow Analysis is a critical phase within this lifecycle, positioned after requirements have been gathered and before logical and physical designs are finalized.

• Systems and Network Services: Includes characterizing services.

• Requirement Analysis:

  • Determining User, Application, Host, and Network requirements.

  • Gathering requirements, service metrics, and performance levels.

  • Characterizing behavior.

  • Establishing performance thresholds and levels.

• Flow Analysis (The focus of this lecture):

  • Identifying Data sources & sinks.

  • Establishing Flow Models, Boundaries, Distribution, and Specifications.

• Logical Design:

  • Technology choice and interconnection mechanisms.

  • Network Management and Security.

• Physical Design:

  • Cable plant design options.

  • Network Equipment placement.

• Addressing & Routing:

  • Creating network diagrams and diagramming worksheets.

  • Establishing routing flow.

  • Developing addressing & routing strategies.

Introduction to Flow Concepts

• Definition of a Flow: Flows are end-to-end information transfers between a source and destination application(s) or host(s) occurring in a single session.

• Flow Composition: A flow consists of a set of application and protocol information sharing common attributes transmitted during a single application session. Common attributes include:

  • Source address.

  • Destination address.

  • Options.

  • Information type.

  • Routing.

• Core Concepts:

  • A flow relates to an end-to-end connection with constant addressing and service requirements.

  • The term "Flow" is used to aggregate service performance characteristics, which are then analyzed and controlled per flow within a Flow Specification (Flowspec).

  • Flow Analysis is used for Capacity Planning (for Best-Effort Services) and Service Planning (for Specified Services).

  • It assists in identifying, sizing, and choosing flows, ensuring each flow has Cumulative Performance Specifications.

The Flow Analysis Process

Flow Analysis provides an end-to-end perspective on requirements, showing how they interact and combine. It offers insights into the necessity for hierarchy and redundancy while guiding interconnection strategies. The process model includes:

1. Establishing Flow Boundaries: Separating portions of the system.

2. Identifying Backbone/Composite Flows: Grouping individual flows.

3. Developing Flow Specifications: Creating a performance requirement document.

4. Identifying Capacity/Service Plans: Determining the necessary network resources.

• Supporting Tools: Flow Characteristics, Flowspec Algorithms, Flow Models, Flow Distributions, Requirement Specifications, and Application Maps.

Categories of Flow

There are three primary types of flows in a network hierarchy:

• Individual Flow: The basic unit of traffic flow for a single application session. Best Effort and Specified Services flows are typically considered separately.

• Composite Flow: A combination of Best Effort individual flows that share the same path, link, or network. These are primarily used in Capacity Planning.

• Backbone Flow: A hierarchical composition of Composite Flows across the network. These indicate the formal hierarchy within the network infrastructure.

• Relationship Formula: Backbone flows are sets of Composite flows, which are in turn sets of Individual flows. $\\text{Backbone} = \\{\\text{Composite} = \\{\\text{Individual}\\}\\}$.

Postal Service Analogy for Flows

• Individual Flow: A single letter sent from one person to another.

• Composite Flow: Neighborhood letters sorted and placed into a single delivery truck.

• Backbone Flow: Multiple trucks traveling between large regional distribution centers.

Flow Elements: Data Sources and Data Sinks

• Data Source: A device or group of devices that generate data for the network to carry. These are usually high-computing devices. Examples include:

  • Servers and Mainframes.

  • Parallel systems and Clusters.

  • Cameras and Video equipment.

  • Medical scanners.

• Data Sink: A device or group of devices that primarily accept or collect data from the network. Examples include:

  • Data storage systems (Tape or Disk groups).

  • Video editing equipment.

  • Specialized display equipment.

Traffic Flow Example (Quantified)

An example university environment shows how different departments contribute to total flow:

• Administration: $50\\,PCs$ and $25\\,Macs$.

• Business and Social Sciences: $50\\,PCs$.

• Arts and Humanities: $30\\,PCs$.

• Math and Sciences: $50\\,PCs$.

• Library and Computing Center: $30\\,Library\\,Patrons\\,(PCs)$ and $30\\,Macs/60\\,PCs$ in the Computing Center.

• Specific Application Flows:

  • App 1: $108\\,Kbps$.

  • App 2: $60\\,Kbps$.

  • App 3: $192\\,Kbps$.

  • App 4: $48\\,Kbps$.

  • App 5: $300\\,Kbps$.

  • App 6: $200\\,Kbps$.

  • App 7: $400\\,Kbps$.

  • App 8: $1200\\,Kbps$.

  • App 9: $80\\,Kbps$.

• Server Farm Connection: $10\\text{-Mbps}$ Metro Ethernet to Internet.

Flow Modeling

Flow models help identify flows via directionality and hierarchy. There are four major models:

• Peer to Peer: Users and applications have similar communication requirements. Flows are equally likely between any hosts. This is the default model.

• Client-Server: Flows are asymmetric, favoring the direction toward the clients. The server acts as the primary data source.

• Cooperative Computing: Multiple applications work together and share info. It introduces hierarchy because work must be managed. Flows exist between clients/servers and servers/managers.

• Distributed Computing: The most specialized model based on the relationship between task managers and computing nodes.

  • Coupling: Can be "closely coupled" (frequent transfers) or "loosely coupled" (little communication).

  • Granularity: "Coarse-grained" (tasks dedicated to single nodes, usually loosely coupled) or "Fine-grained" (tasks subdivided based on parallelism, usually closely coupled).

Flow Boundaries (FB)

Flow boundaries represent separations between large aggregated portions of the system where flow consolidation naturally occurs.

• Common Geographic Boundaries:

  • LAN/WAN and LAN/MAN.

  • Campus/Campus.

  • Building/Building.

  • Floor/Floor (Campus, building, and floor are subsets of a LAN environment).

• Logical/Traffic Boundaries: Separation based on logic rather than geography:

  • Backbones (where several flows transit).

  • Flow Concentrations (convergence points like an NAP).

  • WANs (where service providers are used).

  • Specialized Areas (specific service requirements).

Flow Distributions (FD)

Flow distributions help locate backbone flows by distinguishing between localized flows (staying in one region) and transit flows (crossing boundaries).

• Traditional 80/20 Rule of Thumb: Historically, $80\\%$ of flow stays within the LAN, and $20\\%$ transits the WAN. This means WAN capacity is roughly a quarter of LAN capacity.

• Modern Shifts: Remote computing is changing these ratios to $50/50$ or even $20/80$ (Distance Independent Computing).

Flow Specification (Flowspec)

Flowspec combines application requirements into a performance document. Units are typically categorized by complexity:

• Unitary: For capacity planning of Best-Effort flows only. No specified flows.

• Two-Part: Contains both Best-Effort and Specified Flows. Builds on Unitary info.

• Multipart: Provides high detail on individual components of Specified Flows.

The Flowspec Algorithm

1. List characteristics of each flow.

2. Combine Reliability, Capacity, and Delay characteristics using the algorithm.

3. Condition 1: Only Capacity is used for Best Effort (BE) calculations (BE cannot guarantee reliability or delay).

4. Condition 2: Use all characteristics for Specified Flows. Perform calculations to maximize overall performance.

5. Condition 3: Guaranteed delay and reliability requirements are used individually.

6. Condition 4: Capacities generated are "Baseline" capacities and do not reflect performance modifiers.

Formula/Logic for Flowspec Types

• Unitary: Determined by summing each flow $SC_{BE}$.

• Two-Part:

  • Capacity Best Effort: Same as Unitary.

  • Specified Environment: Sum of capacities ($SC_{DET}$) plus Maximum Reliability ($R_{DET}$) and Minimum Delay ($D_{DET}$).

• Multipart: Extends the Two-Part spec by adding the Guaranteed Environment ($C_i, R_i, D_i$).

Questions & Discussion

Topic: Identifying Data Sources and Sinks

• Scenario A: A storage device receiving streaming video from a camera.

  • Role: The camera is the Source; the storage device is the Sink.

• Scenario B: A video editing device using video from (a).

  • Role: The storage device becomes the Source; the video editing device is the Sink.

• Scenario C: A Web server and its clients.

  • Role: The Web server is generally the Source (sending data), and clients are Sinks; however, in requests, the client acts as a Source. It depends on the direction of the specific flow.

• Scenario D: A storage disk farm.

  • Role: Acts as a Sink when saving data and a Source when data is retrieved from it.