Chapter 25: Optimizing Data Network Bandwidth

Overview

Networks have undergone tremendous changes over the past few decades.  The special-purpose terminal has given way to the general-purpose desktop computer.  All users need to know is the name of the server where desired information resides and how to connect to it.  The typical network of today is a network of LANs.  They require intelligent design, constant monitoring, and necessary bandwidth.

Changes Affecting Data Networks

LANs are typically connected through frame relay or a virtual private network that creates a secure tunnel through the Internet or the carrier’s IP network.  In recent years, there have been dramatic changes affecting data networks.  These changes include:
·        Users don’t distinguish the boundary between LAN and WAN.
·        Digital circuits have practically replaced analog.
·        End users want their computers networked with access not only to the corporate mainframe, but also to file servers, e-mail, desktop video, etc.
·        Applications require more and more bandwidth.
·        Bandwidth requirements are increasingly difficult to predict.
·        Corporate information is becoming widely distributed among numerous servers.
·        Organizational charts are flattening and are increasingly volatile.     

Internetworking Design

Circuit switching behavior is predictable and controllable up to the network’s capacity.  Legacy data networks, which are controlled by a central host, also behave predictably and can be modeled by formulas that follow the statistical distribution of traffic.      

When analyzing an Internet design, or upgrading a network, there are questions that must be answered to characterize traffic: 
·        Where are the communicating nodes located?
·        Which nodes are the focal points of the bulk of communications?
·        What is the nature of the traffic?
·        What is the statistical distribution of traffic? 
·        What are the periods of peak traffic flow?
·        What kind of response time is expected?

Traffic flows can be segregated into four categories based on two criteria.  First, determine whether the traffic flow is symmetric or asymmetric; second, determine whether the traffic flow is steady or bursty?  Steady and symmetric characterize the traffic of voice and video.  Symmetric and bursty are characteristic of server-to-server and E-mail.  Asymmetric and steady characterize server backup, while asymmetric and bursty are associated with traffic of the host-terminal, file transfer, web surfing, and client-server.

The main difference between service management in voice and data networks:
·        Voice - managing circuits
·        Data – managing raw bandwidth (bandwidth is expensive)

Frame Relay Design

Frame relay is a data communications service that transports frames of information across a network to one or more points.  It is often used for the headquarters and branch office network.  In most cases, the headquarters will need T-1 access.  The branch office access circuits depend on the number of users and the applications.  Small offices with fewer than 20 users may get by with 56-kb/s access.  Larger offices likely will require T-1 access.  Cost is based on three elements:
·        Committed information rate (CIR)
·        Access circuit
·        Post speed

Virtual Private Network Design

There are two of types VPNs: remote access and site-to-site.  Design of site-to-site networks is similar in concept to frame relay.  Architecture is the same, but there are significant differences in security and bandwidth considerations.  A site-to-site VPN has the advantage that the network is implemented over a structure that is in place for Internet access.

Network Quality of Service

Not all traffic on a data network is treated equally.  Demanding traffic is prioritized over less important traffic.  The priority may be a function of what type of traffic it is or it may be a function of who is sending it; for example, customers or internal users.  The methods of prioritizing traffic generally fall into the following categories:
·        Bandwidth reservation
·        Queuing
·        Traffic shaping
·        Flow control

Flow control works well for data, but not with delay-sensitive applications.  The network handles this by identifying a flow and either reserving bandwidth as with RSVP, or tagging the flow to shortcut the routing process.

The first task in quality management is classifying the traffic.  This may be based on the following variables:
·        IP address
·        Time-of-day
·        Protocol type
·        Port
·        Class of user

Policy Management

In a data network, a policy is a set of rules that controls traffic distribution through network.  The policy must be implemented in a server that can prioritize, filter, and block traffic based on such variables as who is sending it, the time of day, what kind of traffic it is, and network congestion.

Another issue with policy management is where the information resides.  File servers implement policy by establishing rights and permissions for different directories and files.  A policy would establish who has permission to access personnel files.  Ideally, the policy would regulate the servers, the network, the PBX and maybe even the internal security system.

The trend in the industry today is to use lightweight directory access protocol (LDAP) as the central repository for technical information about an individual.

Policy management systems must have protocol such as COPS to link the policy servers and the PEPs.  Policy rules must also be prioritized so that if a conflict occurs, the system knows which rule prevails.  Policy management systems consist of the following:
·        Policy server
·        Policy decision points
·        Policy enforcement points

Traffic Management

Traffic management devices identify traffic types and prioritize it to make sure the most critical traffic gets through.  Traffic shaping is the process of containing traffic within prescribed limits.  The following types of shaping are common among bandwidth managers:
·        Latency enforcement
·        Bandwidth enforcement
·        Mixed-class traffic handling

Load balancers are used to combine the capacity of multiple lines into a logical path.  Load balancers fall into the following categories:
·        Software applications
·        Hardware
·        Load-balancing appliances

Bandwidth and Policy Management Selection Considerations

The following issues and questions are just some that should be addressed in selecting bandwidth and policy management products:
·        Ease of implementation
·        Growth capability
·        Is policy server centralized?
·        Does bandwidth management need to address functions such as security, firewall, and VPN?
·        How do vendors download and enforce policies?
·        How does the vendor’s device classify traffic?
·        What is the vendors packet-forwarding rate without loss?
·        How does vendor define service classes?
·        What vendor measuring and monitoring tools are used?
·        Does the vendor system implement rate enforcement?