Selecting the right tool depends on who will use it and why.

Discover several features to look for when you choose your next network design tool.
Arnold W. Bragg

Which Network Design Tool Is Right for You?

Inside

eve all seen advertisements for commercial network design tools. They usually show colorful workstation and server icons suspended like insects in a spider web of switches and routers, all neatly superimposed on a map of Northern California. But network design is more than dragging, dropping, and connecting icons to build a graphical network infrastructure. Network design is challenging, requiring designers to balance user performance expectations with network-resource costs, capacities, capabilities, and use levels. Network operators want to keep resource utilization high and costs low. But then the resulting design might produce unacceptable service levels when congested links and undersized routers cause too much delay or too many dropped packets.You need a design that will efciently use your network resources and effectively serve your users.And you want a network design tool that can produce the design you want. Networks are a complicated mix of applications, protocols, device and link technologies, traffic ows, and routing algorithms. There may be tens of thousands of feasible congurations, each with different performance attributes and costs. Can network design tools help you nd the combination that is right for your organization? Yes, but its important to use the right one for the job.

Other NetworkRelated Tools

WHAT IS A NETWORK DESIGN TOOL?

Network design tools help IT professionals sift through con-

icting features, trade-offs, and thousands of reasonable design alternatives. Is the designer building a new network or maintaining and troubleshooting an existing one? Will proposed changes in the network handle the anticipated trafc demand without violating service-level agreements? Does the design require wireless or satellite links? Is the tool suitable for presentations to customers, and can it provide fast answers to their what-if questions? Is the designer testing an experimental protocol or a novel routing algorithm? Tools analyze (visually, mathematically, or by simulation) what is happening in the network and predict future behavior. Nearly every general-purpose network design tool works the same way.The designer either uses a drag-and-drop graphical editor to create or modify a network topology or imports the topology directly from a network management tool like HP OpenView. Most tools have extensive libraries of device and link models, so building an Ethernet local area network (LAN) or IP-routed backbone network is straightforward. Next, the designer specifies (or confirms) the attributes of each device, link, or applicationbit rate, traffic load, device throughput (in bits or packets per second), transaction rate, routing protocol, and so onand simulates the networks operation. Simulation runtimes vary from several minutes to several days, depending on the required level of accuracy and detail in the results. Most tools can display results graphically at the end of the simulation run, and many use animation or dynamic utilization bars during the course of the simulation to highlight bottlenecks and traffic flows. Nearly every tool can identify and
September October 2000 IT Pro 23

1520-9202/00/$10.00 2000 IEEE

Authorized licensed use limited to: University of Pittsburgh. Downloaded on January 6, 2010 at 16:31 from IEEE Xplore. Restrictions apply.

NETWORK

DESIGN

Figure 1. Opnet Technologies IT DecisionGuru portrays a network of interconnected backbone IP routers.

summarize network congestion, load, lost packets, failure and error conditions, and response times. Network design tools also let you evaluate different scenarios. For example, suppose you need to evaluate the trafc load introduced by three different routing protocols on a portion of an existing wide area network (WAN). Figure 1 shows how one tool portrays the network as a set of interconnected backbone IP routers, each directing traffic to and from one or more abstract subnetworks. In this example, the border gateway protocol (BGP) generates slightly less routing trafc than the open-shortest-path-rst (OSPF) protocol, and both generate far less routing trafc than the routing information protocol (RIP). Conguring the three scenarios in this example may be as simple as selecting a different value for the routing protocol attribute from a single pull-down menu. Quantitative network design tools produce more accurate and defensible results than qualitative methods or rules of thumb. A network operator may have expected BGP and OSPF to generate less routing trafc than RIP but couldnt predict the relative performance of OSPF and BGP for this specic topology. Network design tools are also valuable for predicting performance problems rather than merely reacting to them when they arise. Simulation is not the only solution. Several tools rely partly or solely on mathematical analysis and queuing theory. Analytical tools provide solutions more quickly than simulation but dont always achieve the required accuracy.
24 IT Pro September October 2000

WHO USES NETWORK DESIGN TOOLS?

The ideal network design tool can mean different things to different people. But whether youre a network designer, network manager or engineer, sales or marketing manager, or member of the R&D staff, you probably want intuitive graphical interfaces that resemble computer-aided design tools. Distances are important in LANs and WANs, so the tool must recognize and compensate for user-dened scales.You want several levels of operational abstraction, so you can model a LAN as 37 distinct packetgenerating devices at one extreme, and as a cloud raining 10,000 packets per second at the other. You also want to test whether your proposed design or change can handle the workload with acceptable performance. However, depending on your situation, concepts of the ideal network design tool can vary greatly.

Sales and eld-services staff

You sell network hardware or provide network services and want to show customers reasonably accurate representations of how a product, service, or technology will improve their network and support their business case. You want an intuitive tool that runs on a laptop computer and that you can master in just a few days. You need fast execution speed (tens of seconds per scenario rather than tens of minutes) and extensive presentation features (such as trafc animation, graphical indications of service-level performance, reports, diagrams, charts, and graphs).

Authorized licensed use limited to: University of Pittsburgh. Downloaded on January 6, 2010 at 16:31 from IEEE Xplore. Restrictions apply.

Network managers and engineers

You operate the networks, troubleshoot and solve performance problems, and ensure compliance with servicelevel agreements.You try to gauge how proposed changes will affect cost, performance, capacity, and availability before the changes go through. Changes typically include introducing new routing protocols, adding new devices and links, supporting new applications such as those for enterprise resource planning (ERP) or e-mail, upgrading servers, and changing service-level agreements. If the existing network is large, you must import topology and trafc data from other tools. The set of alternatives is usually enormous, so you need to evaluate scenarios in tens of minutes rather than hours.

conrm that service-level agreements can or cannot be met; predict the response times of applications before deploying them; predict where bottlenecks and application response time delays will appear rst; predict how many users an application can handle; and present results via animated utilization bars, online graphical statistics and counters, diagrams, reports, charts, and graphs.

Not-so-easy problems

General-purpose tools only partially solve some problems. Here are some of the more common ones. Determine the best performance-cost ratio. Lets say you want to decide which devices, LAN or WAN technologies, Network designers You specify and build new networks or overhaul exist- or routing protocols give the best performance at the lowing ones.You try to reduce design time and improve design est cost. Exploring alternatives is easy, but the user must accuracy. You want your designs to meet performance specify the scenarios and evaluate the costs and results, requirements without overbuilding. And you want to be usually one by one. Some tools generate a bill of materials able to identify potential bottlenecks and overloads. for the network devices, but few general-purpose tools facDesigners need an extensive library of link technologies, tor in communications costs. Estimating costs is challenging because telecommunications devices, architectures, and protocols One of the most tariffs vary widely by location and to build or upgrade the network, and you need tools to accurately challenging decisions is type of servicefor example, leased lines versus frame relay. predict its performance. Provide accurate solutions in secdetermining how much onds. You must trade off execution Researchers accuracy to sacrice speed and accuracy. Speed comes To reduce development costs and from high levels of abstraction, as risks, you test the effects of new or for speed. when a tool models a large LAN as modied protocols, devices, archia cloud generating an aggregate tectures, component designs, and trafc ow.To test a new component or protocol, you need trafc models in the lab or on the workbench. You need complete control of simulated behavior at the programming to simulate every packet passing through it. To gure out language level, and you want the language to provide a rich whether you can add five workstations to that overburset of special-purpose modeling functions.You typically sim- dened LAN on the fourth floor, using aggregate flows ulate discrete events (packets transiting a router, protocol makes sense.There is some middle ground, though. Some retransmissions, and so on), and mimicking actual behavior tools let you simulate part of the network in detail while can take billions of events.You prefer accuracy over speed. modeling trafc in the background or in other parts of the network as aggregate ows. One of the most challenging decisions is determining how much accuracy to sacrice WHAT GENERAL-PURPOSE TOOLS DO Many network design problems are relatively easy to for speed. Realistically model an existing network. Realistically solve. Nearly every general-purpose tool will modeling a large network can be difcult if each device, model LAN, frame relay, IP, and asynchronous transfer link, and trafc load must be manually congured. Most tools can import a network topology from a netmode (ATM) segments and traffic using stock library work management suite like HP OpenView, or discover models; and import the topology with a tool based on the simple predict link and device throughput and utilization; network management protocol (SNMP). Some tools show which links and devices are over- and underused; import trafc data or proles from a network sniffera show how (un)balanced loads affect links and devices; device that monitors and characterizes the trafc passing show the effects of link or device failures; show how changes in topology, device upgrades, or over a communications link. Others have rapid conguration wizards to quickly build large networks. A few have expansion will affect performance; identify the applications or protocols contributing most databases with tens of thousands of voice, video, and data devices that you can drag into your design window. to network congestion and delay;
September October 2000 IT Pro
Authorized licensed use limited to: University of Pittsburgh. Downloaded on January 6, 2010 at 16:31 from IEEE Xplore. Restrictions apply.

25

NETWORK

DESIGN

Capturing trafc behavior is usually more important than coarse estimates provide insight and reduce the number modeling devices in minute detail. If you need to import of scenarios you need to simulate. topologies or trafc proles, make sure the tool can do this. Simulate models with thousands or tens of thousands of Stress test the network. Stress testing means ratcheting devices. This problem is common when importing topoloutilization levels up over 90 percent. Simulating every gies from network management systems. Discrete-event packet owing through a congested network leads to very simulators may run out of computing resources when evallong simulations. But watch out for analytical tools that uating very large networks. Consider using a tool with an blithely report link utilization above an impossible 100 analytical engine. Analytical tools use steady-state queupercent or tools that do not accurately reect how protoing theory formulas and mathematical modeling techcols like TCP respond to packet loss at high loads. Select niques to produce quick solutions. They dont capture a tool that can distinguish between the load offered to the nuances of protocol behaviors, but most provide good rstnetwork and the load actually carried by the network. order approximations. Analytical tools may be the only Balance power and ease of use. reasonable approach for very large Some tools offer Powerful tools have steep learning networks. curves. Power comes from the ability Model mobile or wireless compoboth full-featured to precisely configure or customize nents. Mobile and wireless networks designer versions device and protocol behavior. Some have a unique set of problems that most tools offer both full-featured designer and less-expensive general-purpose tools do not address. versions and less-expensive runtime Uneven terrain,satellite trajectories,3D versions. The latter usually have condynamic positioning, and fading signals runtime versions. figurable attributes (such as link require special-purpose tools. Some speed). vendors offer these as separate add-on Conrm that devices are connected correctly. Most tools modules. Dont buy them unless you need them. ensure that devices and links are correctly matched, and Simulate every nuance of link and network protocols. some give a warning if, for example, a router is not comOne of the most difficult problems in selecting a tool is patible with the optical link its connected to. But few tools deciding what level of detail you require. Some tools cost conrm that LANs are within their maximum lengths and ve to 100 times more than others, so this is an important that the network correctly compensates for delays over decision. Decide whether you need to model behavior at long-distance WANs. For example, if a tool tells you there the programming language level. If so, make sure the tool is zero propagation delay on that Boston-to-Seattle link, has a rich set of functions and procedures, and that you you have a problem. can modify its stock models. If your goal is to simulate Provide statistically useful output. Most tools simply some variant of TCP, you can save lots of effort by starting compute average delay and utilization, but youll want to with the stock TCP model and customizing it. know how high utilization was between 2 and 4 p.m., and how much trafc exceeded the maximum delay promised A SAMPLING OF GENERAL-PURPOSE TOOLS by the service-level agreement. If you need to plot time If you need a general-purpose network design tool, conseries, histograms, or condence intervals, make sure the sider these first. Most have earned at least one editorstool youre considering can do the job. choice or best-in-class award. They all have a graphical toolkit for building or importing network topologies and trafc ows.They all predict or portray a networks behavTough problems ior or performance in some way. And they all have an Some problems are quite difficult to solve. Heres a extensive set of reports, diagrams, charts, and animated short list. displays. Each tool supports all the common link and transOptimize network design or performance. Network port protocols, LAN and WAN architectures, routing prodesign tools do not claim to provide optimal solutions. tocols, and devices. Most support some level of customiOptimization is very difficult, and objectives often conzation through simulation attributes. ict. Users and network operators have different sets of requirements and constraints. Automate the what-ifs. Most designs have more options Analytical Engines NetRule than a tool could evaluate in a lifetime.The usual approach and NetRule Viewer is to identify a few feasible alternatives and test them. Analytical Engines advertises NetRule (http://www. During the evaluation, other alternatives may emerge. In netrule.com) as an analytical tool for planning network theory, tools could automate this process. In practice, they changes. The company has clearly positioned NetRule as a usually do not. Is it better to mathematically evaluate a mid-level tool optimized for practical use on real networks few hundred designs with moderate accuracy or to simuand less suited for research and development of new protolate a dozen with great precision? It depends. Sometimes cols or component designs as stated in the product descrip26 IT Pro September October 2000
Authorized licensed use limited to: University of Pittsburgh. Downloaded on January 6, 2010 at 16:31 from IEEE Xplore. Restrictions apply.

tion. The company intends this tool for use by network Figure 2. Modeling a wide area network with NetRule. engineers involved in network planning. Figure 2 shows the components available to build a WAN in NetRule. NetRule is written in Java and runs on the thirty or so platforms that have Java virtual machines. It also uses analytical methods instead of discrete-event simulation, so you can usually evaluate its models in tens of seconds. NetRules developers believe that simulation is too complex, too slow, and too expensive for practical network planning. The analytical engine uses steady-state queuing theory formulas, mathematical modeling, search algorithms, component-level simulation, and rule-based inference tools. Advertisements claim evalua- agement and monitoring tools and, if a user requires more tion times rarely exceed 10 seconds, even for large networks. precision, export topologies directly to Comnet III. NetRule can model and stress test very large networks EcoProler is a specialized tool for assessing the effects (1,000 to 10,000 nodes) because runtimes dont depend on of new ERP applications on network performance, as well the number of packets transiting the network. as the opposite problem of assessing how adding users or NetRule also has a viewer for read-only access to net- network upgrades affects existing ERP performance. Some work modeling results. Users can replay analyses but can- ERP applications have rather stringent resource requirenot alter them. ments, so you can use tools like EcoProler to assess an applications performance before deploying it.

Compuwares Comnet III, EcoPredictor, and EcoProler

Compuware (http://www.compuware.com) acquired several excellent tools from CACI International in 1999.Two of them are mature products with large installed bases. Comnet III is a high-end design and discrete-event simulation tool. Designers can create hierarchical network models using a drag-and-drop tool palette, or they can import topology and traffic data from several network management tools. Tightly integrated with other Compuware tools, Comnet III has an extensive library of devices and protocols. It also has optional add-on modules for circuit-switched traffic (voice, video, satellite), distributed applications, and mobile or wireless networks. EcoPredictor is a mid-level analytical performanceprediction and capacity-planning tool, advertised as fast, scalable, easy-to-use, and ... designed for day-to-day use. The tools most interesting features are reports and alarms that identify a networks top contributors to congestion. It also locates bottlenecks and excess capacity. EcoPredictor can import data from several third-party man-

NetCrackers Designer and NetCracker Professional

NetCracker Technology (http://www.netcracker.com) has two network design products: NetCracker Designer and NetCracker Professional. Designer is a low-cost, lowend tool for designing, verifying, documenting, and visually analyzing networks via animation. It can import topologies from Visio and OpenView, and automatically conrm device connectivity and compatibility. NetCracker Designer supports the most common network and routing protocols and media. Its most interesting feature is a database of more than 25,000 network devices. Each device has several dozen attributes (supported media, protocols, port congurations, latency, bandwidth, price, and so on). NetCracker Designer can use the network topology and device database to generate a bill of materials. Database updates are available by subscription. NetCracker Designer simulates static trafc ows by animation but does not simulate or analyze performance. NetCracker Professional reads Designer files, simulates
September October 2000 IT Pro 27

Authorized licensed use limited to: University of Pittsburgh. Downloaded on January 6, 2010 at 16:31 from IEEE Xplore. Restrictions apply.

NETWORK

DESIGN

Figure 3. Modeling a circuit-switched network with NetCracker Professional.

ance attributes. You can specify trafc loads in several ways,including by application type.

Opnet Technologies Modeler, IT DecisionGuru, and Netbiz

Opnet Technologies (http:// www.opnet.com) also has a family of mature tools and a large installed base. Modeler is a more expensive, high-end product used mostly by network R&D engineers. It can very precisely model protocols,devices,and behaviors using a nite-statemachine paradigm, C/C++ language features,and about 400 special-purpose modeling functions. Modeler has optional add-on modules for radio and satellite modeling, multivendor import, and service-level prediction. Originally a discrete-event simulator, Modeler now supports hybrid simulations, which combine discrete-event simulation and analytical modeling. It can also run a simulation in parallel over several CPUs. Both hybrid and parallel simulations can significantly reduce simulation runtimes. IT DecisionGuru is a mid-level analytical tool for network architects, managers, and planners (see Figure 1). Advertisements say that its easy to use and that you can become procient with it in just one day. IT DecisionGuru shares Modelers vast protocol and device library. Opnet Technologies developed it jointly with HP (which may explain its impressive capabilities for profiling applications and importing and digesting packet trace information). IT DecisionGuru can also predict and display the impact of changes on service-level agreements. Netbiz is for sales, professional services, and eld services staff. It provides a rules-based environment for network design that you can customize to your organization. It has excellent presentation and configuration features and boasts a minimal learning time. It targets companies selling networking services or equipment. Figure 4 is a subjective summary of the intended audience and skill requirements for these general-purpose design tools.

Figure 4. Intended audience and required skill for 10 generalpurpose network design tools.
More
Mode ler

Comne

t III

Skill required

EcoPredictor and IT DecisionGuru NetRule

EcoP
Netbiz

ro

filer
Designer NetRule Viewer

Profess

ional

Less Sales or field-service staff

Network Network managers designers or engineers Intended audience

Researchers

behavior and performance, and generates reports, diagrams, charts, and online dynamic graphs, as shown in Figure 3. Animations are impressive, with graphical bars showing utilization, packet loss, load, and other perform28 IT Pro September October 2000

OTHER TOOLS
Several network performance tools, application and

Authorized licensed use limited to: University of Pittsburgh. Downloaded on January 6, 2010 at 16:31 from IEEE Xplore. Restrictions apply.

Other Network-Related Tools

Network performance tools
Mimic (Gambit Communications, http://www. gambitcomm.com): SNMP (simple network management protocol)-based enterprise management tools for network management developers and users. Chariot, Pegasus (Ganymede Software, http://www. ganymede.com): A tool intended for managing, testing, monitoring, and troubleshooting network performance. NetClarity (LANQuest, http://www.lanquest.com): Network performance management and diagnostic tools for load balancing, capacity planning, and service-level management. NetMaker (Make Systems, http://www.makesystems. com): Network planning, network engineering, and trafc analysis. Link Analyst (Network Instruments, http://www. networkinstruments.com): Graphical network mapping, route analysis, quality-of-service management, troubleshooting, and up-time monitoring. NetSpec (University of Kansas, http://www.ittc. ukans.edu/netspec): Trafc generation tools for network experimentation, passive measurement, and performance testing. optimal.com): Visualizing, reporting, and tuning ecommerce, ERP, and Web-based applications over wide area networks. WebLoad (RadView Software, http://www.radview. com): Simulating and conrming real-world trafc to test Web applications performance and scalability requirements.

Special-purpose tools
SystemView (Elanix, http://www.elanix.com): Designing and simulating analog, wireless, cellular, and other communications systems and networks. Odyssey (Logica, http://www.logica.com/offerings/): Design tools for radio-based networks. Visio 2000 (Microsoft, http://www.microsoft.com/ ofce/visio): Network drawing and diagramming tool. Simulink (The MathWorks, http://www.mathworks. com): Digital signal processing and control system component design. Telsoft (Telsoft International, http://www.telsoft. com): Extracting and presenting real-time network activity, trafc, and routing data; primarily for telecommunications networks. VPI Suite (Virtual Photonics, http://www.virtu alphotonics.com): Bandwidth design tools for physical layer components, and access and transport networks. NetSuite (Visionael, http://www.netsuite.com): Discovering, validating, and designing complex networks. VisSim (Visual Solutions, http://www.vissim.com): Modeling and simulating analog, digital, and mixed-mode components and systems. NPATNetwork Planning and Analysis Tool (WANDLWide Area Network Design Laboratory, http://www.wandl.com): High-end tools for network design, capacity planning, failure analysis, and network management; includes a sophisticated database of tariffs.

Application and business performance tools

Total Control (Lightspeed Systems, http://www. lightspeedsystems.com): Traffic load, quality of service, rewall, and server monitoring and control. VitalSuite 7 (Lucent Technologies, http://www.ins. com): Network application and business performance data, custom reports, and views. S3 (NextPoint Networks, http://www.nextpoint. com): Managing networks, e-commerce, IP services, and applications designed to support the business objectives of the enterprise. Application Expert (Optimal Networks, http://www.

business performance tools, and special-purpose tools have network design features. See the Other NetworkRelated Tools sidebar for more information.

DEFINING YOUR NEEDS

Do you need a special-purpose tool, or will a generalpurpose one suffice? If you design and test components for optical networks, then that six-figure, one-of-a-kind optical network design tool you saw at the trade show

might just be what you need. As noted, some performance monitoring tools have network design features, so you may already own a low-end network design tool. If you just want to draw network topologies, stick with Visio. How much are you willing to spend? How fast must the tool evaluate scenarios? How much time can you invest to master the tool? If you think you need a general-purpose network design tool, what type of user are you?
September October 2000 IT Pro 29

Authorized licensed use limited to: University of Pittsburgh. Downloaded on January 6, 2010 at 16:31 from IEEE Xplore. Restrictions apply.

NETWORK

DESIGN

Table 1. Consider which features best support you, based on the type of user you are.
Type of user Sales or eld-services staffer x Network manager or engineer x Network designer x

Feature Represent resource capacity and bandwidth Measure, estimate, and report utilization; network congestion; load; delay; lost packets; response time; device latency; or availability Identify bottlenecks Support exploration of alternatives Perform sensitivity analysis Specify trafc and server workloads Distribute load across complex topology Handle high-load and other stress tests Model failure and error conditions Predict when and where the network will develop performance problems Predict which device, link, or subnetwork will reach capacity rst Identify applications and protocols causing congestion Ensure service-level conformance Identify nodes and links close to exceeding user-dened thresholds Identify underutilized resources Predict how many users an application can support

Researcher x

Prevalence Common

x x x x

x x x x x x x x

x x x x x x x x x x x x x x x

x x x x x x x x x x x

Common Common Common Uncommon Common Uncommon Uncommon Common Common Common Common Uncommon Uncommon Uncommon Uncommon

x x x x x x x

x x x x x x x

Features
In the table, I list various features to help you decide what kind of network design tool you need. (For a far more extensive table, see http://computer.org/itpro/it2000/pdf/ f5023x1.pdf on the IT Pro Web site.) Depending on what you do, you can use this table to select features you consider important or at least something you should be aware of.The table also indicates how prevalent these features are. Reconsider your execution speed requirements. Execution speeds span four orders of magnitude, from tens of seconds to days. If you really need to simulate every packet, tune a routing protocol, or tweak TCP, then you absolutely need a discrete-event simulator.To congure a network design while the customer waits, or to simulate a very large network, you probably need an analytical tool or a tool streamlined for eld use. Ask the vendor about hybrid solution methods and about techniques that simulate abstract trafc ows rather than individual packets. If
30 IT Pro September October 2000

you purchase a mid-level or high-end product, invest in two to five days of training to learn techniques for optimizing execution speed. Discrete-event simulators can deliver very precise results in tens of minutes if the user can specify (and the tool can handle) less-critical background details at higher levels of abstraction.

Price
Once youve settled on the feature set, reconsider your budget. Prices for general-purpose tools vary from several hundred to more than fty thousand dollars. Most vendors have several levels of products: high-end tools for developers; runtime and lite versions for the sales and eld services staff; and add-on modules for specific applications like wireless networks.

Authorized licensed use limited to: University of Pittsburgh. Downloaded on January 6, 2010 at 16:31 from IEEE Xplore. Restrictions apply.

Vendors can give you good advice if you talk to them. A good t is mutually beneficial. Dont buy a five-figure tool if a four-figure tool is sufficient, and dont settle for a four-gure tool if you really need that ve-gure tool.

How to Reach IT Pro

Writers
We welcome submissions. For detailed information, write for a Contributors Guide (itpro@ computer.org) or visit our Web site: http://computer.org/itpro.

etwork design tool vendors often change company or product names. Companies are bought and sold. The tool you evaluated in 1998 might not be the tool you get today. Ask for a demo version. Check the pedigree.And dont buy add-on modules until you need them; you may decide you dont. As for future trends, hybrid engines make sense. Expect more vendors to support a mix of discrete-event and analytical solution methods, with tuning knobs to make the speed/accuracy trade-offs easier.Also expect some network and application performance tools with network design features to move into the mainstream, particularly at the low end. Expect vendors to offer more addon options, because integrated add-on modules keep network design tools affordable. Finally, techniques from artificial intelligence and operations research might go far in automating the what-ifs.Watch for better ways to gain insight about bottlenecks and congestion points, and evaluate competing design scenarios. I

News Ideas
Contact Lee Garber at lgarber@ computer.org with ideas for news features or news briefs.

Products and Books

Send product announcements to [email protected]. Contact Jason Seaborn at jseaborn@ computer.org with book announcements.

Arnold W. Bragg is a senior member of the technical staff at Fujitsu Network Communications Inc. Contact him at [email protected].

Letters to the Editor

Please provide an e-mail address or daytime phone number with your letter. Send letters to IT Pro Letters, 10662 Los Vaqueros Circle, PO Box 3014, Los Alamitos, CA 907201314; fax +1 714 821 4010; itpro@computer. org.

Becoming a speaker for one of the Computer Societys lecture programs allows you to  present ndings from your latest book or research project  sharpen your presentation skills  experience new parts of the world  enhance the Societys reputation and your own as a technical leader Chapter Tutorials Program One-day seminars on specialized topics Distinguished Visitors Program Conference tutorials or stand-alone sessions

SPEAK

UP!

Magazine Change of Address

Send change-of-address requests for magazine subscriptions to [email protected]. Make sure to specify IT Professional.

Missing or Damaged Copies

If you are missing an issue or received a damaged copy, contact [email protected].

Reprint Permission
To obtain permission to reprint an article, contact William Hagen, IEEE Copyrights and Trademarks Manager, at [email protected]. To buy a reprint, send a query to [email protected] or a fax to +1 714 821 4010.

For more information, see computer.org/chapter/programs.htm or contact [email protected]

September October 2000 IT Pro
Authorized licensed use limited to: University of Pittsburgh. Downloaded on January 6, 2010 at 16:31 from IEEE Xplore. Restrictions apply.

31

Features to Consider in a Network Design Tool

The following table lists various features to help you decide what kind of network design tool you need (see A.W. Bragg, Which Network Design Tool Is Right for You? in the Sept./Oct. 2000 issue of IT Professional). Depending on what you do, you can use this table to select features that you consider important or at least something you should be aware of. The table also indicates how prevalent these various features are.

Consider which features best support you, based on the type of user you are.
Type of user Sales or eld-services staffer x x x x Network manager or engineer x x x x x Network designer x x x x x x

Feature User interface Intuitive CAD-like graphical interface Drag-and-drop tool palette for building or modifying models Network topology overlays map, grid, descriptive background Standard and custom maps, grids, backgrounds Model behavior accurately reects map distances and grid scales Modeling paradigm Hierarchical representation (at network, subnetwork, and device levels) with ability to drill down Standard object-oriented constructs (such as those for inheritance and specialization) Multiple levels of detail and/or abstraction Model individual devices and links Create custom device and link models Library of models for links, media, devices, architectures, protocols, applications, packet types, and so on Database of voice, video, and data devices, including attributes and prices

Researcher x x

Prevalence Common Common Common Common Uncommon

x x

x x x x x

x x x x x

Common Common Common Common Common

x x

x x x

x x x

x x x x

x x x x x x x x x x

Common Uncommon

Represent subnetworks in terms of aggregate behavior Accurately model large networks (1,000 to 10,000 elements) Conrm connectivity based on device and media attributes Network topology Create topology via tool palette Create topology via rapid conguration tool Discover and import topology from network management tools (such as HP OpenView) Discover, map, and import topology via SNMP-based tools Import user-specied subset topology (such as those for LANs, WANs, or access networks) Import topology from external le or drawing/presentation tool (such as Visio)

Common Rare Uncommon

x x

x x x x x x

x x x x x

Common Uncommon Common Common Uncommon Uncommon

Authorized licensed use limited to: University of Pittsburgh. Downloaded on January 6, 2010 at 16:31 from IEEE Xplore. Restrictions apply.

Type of user Sales or eld-services staffer Network manager or engineer Network designer

Feature Types of networks supported Support wireline (including point-to-point, multipoint, broadcast, switched, and routed) networks Support optical networks, including those based on optical layer protocols Support wireless (xed- and mobile-node, satellite, radio, cellular, or microwave) networks Mobile nodes positions change as simulation progresses Wireless models reect terrain, fading, multipath effects Trafc models, traces, or proles Generate trafc via common models (such as constant and variable bit rate, or bursty trafc) Generate trafc by application (voice, e-mail, or Web browsing) characteristics Create custom ows based on user-dened trafc proles Import packet traces from network monitors and sniffers Import trafc proles from monitors and sniffers Import trafc traces or proles from text les Dene load via trafc matrix that represents ows between end systems Specify background trafc as a baseline load on network Model trafc as aggregate ow from LAN or WAN clouds Solution methods Integration of design tool with simulation or analytical tool Analytical solution methods (such as queuing theory or mathematical modeling) Discrete-event simulation Hybrid analytical and simulation methods Parallel simulations over multiple CPUs Ability to vary simulation granularity (from simulating every packet to aggregates and ows) Vary speed of solution method Simulation output Real-time graphical indications of performance (such as for delay, throughput, packet loss, link use, load) Display trafc ows via real-time animation Graphical output (diagrams, charts, or graphs) Reports Export results to spreadsheets, les, or other tools Generate bill of materials and estimate material costs Estimate tariffs for communications links Presentation tools for nontechnical audience

Researcher

Prevalence

x x

x x x x x

Common Rare Common Uncommon Rare

x x x

x x x

x x x

x x x x x x x

x x x x x x x x x

x x x x x x x x x

Common Common Common Uncommon Uncommon Uncommon Common Uncommon Common

x x

x x

x x

x x x x

x x x x

Common Uncommon

x x x

Common Uncommon Rare Common Rare

x x

x x

x x x x x x x

x x x x x x x

x x x x x x x

x x x x x

Common Common Common Common Common Rare Rare Uncommon

Authorized licensed use limited to: University of Pittsburgh. Downloaded on January 6, 2010 at 16:31 from IEEE Xplore. Restrictions apply.

Type of user Sales or eld-services staffer Network manager or engineer x x x x x x x Network designer x x x x x x

Feature Analysis tools Customized plots (such as time series or histograms) Probability functions, parametric curves, condence intervals Compute descriptive statistics (such as mean, variance, minimum, maximum, or quantiles) Analyze output via lters (for example, moving average) Overlay plots, charts, or graphs Level of customization Support user-dened protocols and devices Simulate any required behavior at the programming language level Language-level functions for special-purpose modeling features Source code provided for models Integrated debugging tool Ability to import existing code into simulation Ability to export simulation code Functions the tool must perform Represent resource capacity and bandwidth Measure, estimate, and report utilization; network congestion; load; delay; lost packets; response time; device latency; or availability Identify bottlenecks Support exploration of alternatives Perform sensitivity analysis Specify trafc and server workloads Distribute load across complex topology Handle high-load and other stress tests Model failure and error conditions Predict when and where the network will develop performance problems Predict which device, link, or subnetwork will reach capacity rst

Researcher x x x x x x x x

Prevalence Common Common Common Uncommon Uncommon Common Rare Rare Uncommon Rare Rare Rare Common

x x x x

x x x x

x x x x x x x x

x x x x x x x x x x x x x x x

x x x x x x x x x x x

Common Common Common Uncommon Common Uncommon Uncommon Common Common Common Common Uncommon Uncommon Uncommon Uncommon Not applicable Not applicable

x x x x x x x

x x x x x x x

Identify applications and protocols causing congestion Ensure service-level conformance Identify nodes and links close to exceeding user-dened thresholds Identify underutilized resources Predict how many users an application can support Other factors Computing requirements (hardware and operating system) Ease of use, short learning curve Training, technical support, and documentation Availability of less-expensive runtime version or viewer

x x x x

x x x x

x x x

Common Uncommon

Authorized licensed use limited to: University of Pittsburgh. Downloaded on January 6, 2010 at 16:31 from IEEE Xplore. Restrictions apply.

Type of user Sales or eld-services staffer Network manager or engineer Network designer x x x x x x x x

Feature Availability and cost of special-purpose add-on modules and features Annual maintenance fees Tiered versions (for the sales or eld-services staff, and network managers and engineers) Degree of integration with vendors other products

Researcher x x

Prevalence Common Common Common Common

Authorized licensed use limited to: University of Pittsburgh. Downloaded on January 6, 2010 at 16:31 from IEEE Xplore. Restrictions apply.