f your until-now ad hoc accretion of network segments. You will probably also want to test aspects of network capabilities: Testi
ng exercises network components to find their limits.
Simulation uses the network as a base, then goes off into the
what-if world
that's essential for planning. Naturally, designing large networks (see the article "Design and Conquer") involves simulation, and maintaining networks (see the article "See the Big Picture") involves testing. System administrators are in the tricky position of trying to satisfy everyone on their big network. Here are some automation tools that can help administrators deliver satisfaction.
Good Models Make Good Networks
To manage a complex network and avoid system degradation and downtime, it's critical to have a view -- a model -- of what the network looks like. When big nets go down or slow down, companies lose big money. That's where a good model comes in. Like a road map when you're hopelessly lost and late, a detailed network model can simplify and speed solving network hang-ups. Without a detailed network mod
el? You could be hosed before you know what hit you. As Bob Ross, president of LAN Utilities (Pleasant Grove, UT), developers of LAN test tools, says, "The self-perception of administrators is that they deal with problems within 20 minutes, but that's only after they find out about them."
Luckily, there are tools that can help create detailed and useful models of even the most arcane and far-flung nets. PinPoint Software continues to improve ClickNet, its network-diagramming product that allows administrators to customize graphic views of a network. It interfaces with network management packages (including Seagate's LAN Directory for Windows and McAfee's Saber LAN Workstation), can download information (about cable configurations; types of workstations, servers, hubs, and routers; and mission-critical software packages), and can automatically generate a graphic model of the network.
While Visio's Visio Technical 4.0 and Visio Shapes for Network Equipment may provide more powerful drawing capabilities
-- drag-and-drop diagramming, clean interface, and ease in aligning and moving objects without breaking connections -- they can't directly import information from network management packages. One of the more impressive and intelligent network-diagramming tools is Quyen Systems' netViz. Besides powerful drawing capabilities, netViz automatically discovers and displays information about nodes on a Novell network, identifying servers, workstations, and volumes. Automatic discovery does not rely on some earlier (and possibly out-of-date) database of network information. The larger the network, the more critical automatic discovery is for developing an accurate and useful model of that network for testing or simulation.
These diagramming tools ultimately deliver a static model of a system, which can help an administrator design and implement a network, understand the complexity of an existing one, and suggest areas that might become problematic. But nothing beats real-time information displayed graphically. No
vell's ManageWise 2.1 allows administrators to model (and also manage) a network from a central location (the console runs under Windows 95 or Windows 3.1). ManageWise can
discover and map
all network devices and inventory the current hardware and software of workstations and servers, creating a dynamic model of the network. And not just for Novell's NetWare. ManageWise can also integrate with enterprise consoles such as IBM's System View for AIX, HP's OpenView, and Sun's Net Manager. The result is the ability to see at a glance when printers, servers, hubs, and routers die. Just as important, ManageWise tracks and displays network performance and
warns when
system behavior might suggest impending problems. ManageWise supports many snap-in applications, such as DenMac's Alert Pager, which immediately informs net supervisors of problems. That means an administrator on the golf course can receive a pager message with the news that a key server has died -- unpleasant a
s golf rounds go but essential for limiting network downtime.
Good Simulation Means Better Testing
A network simulation builds on a model, then brings it to life. Like a model train layout, a network simulation gets all the make-believe traffic moving, all the imitation lights blinking, and all the artificial crack-ups resulting. This is especially crucial with large networks, since no mere mortal can keep the whole conglomeration straight in his or her mind. With a good simulation, you can spot bottlenecks, buildups, and backups. Simulation is also a valuable basis for testing, since you can try things out on the simulator before flying the real fighter jet.
Unleashing new applications or new changes on an unsuspecting large-scale network can be disastrous. Developers especially face this problem. "That's one thing I've definitely found," says Ross. "In a development environment, usually you're fairly sequestered. It's usually not that busy of a network. So the engineers typicall
y are developing on this quiet net, and then you throw the software onto a production environment and it goes berserk." New applications may also face mysterious and temperamental real-time problems -- sneaky bugs that appear only at certain times of the day or under certain kinds of traffic loads. The hardest part of crushing these bugs is recreating them. To do that, the system administrator has to be able to replicate not just traffic volume but particular kinds of traffic.
So how do you adequately test a client/server application product that has never seen a really ugly network? Build your own ugly network! Chariot, from Ganymede Software, is a tool that can simulate the traffic of client/server applications. Using Chariot, you can force a server to handle many concurrent requests from "clients" artificially made by the simulator -- for a fraction of the cost of doing it live on multiple hardware platforms. CenterLine Software's QC/Sim has other nifty features for creating ugly environments in which
to test distributed software. Operating as an add-on to QC/Coverage, a code-testing product, QC/Sim can fake scenarios that are difficult to replicate in practice, like network failures, limited disk capacity, and other real-life horror stories.
Bringing It All Together
In an ideal world, you would buy your whole network from a single vendor, with one perfect modeling and simulation product. But in reality, your huge network has probably snowballed until it seems to have a life of its own. "Larger organizations with multiple deployed offices don't go to Ford and buy," observes Bill Gillman, vice president of operations at Gentner Communications Corp. (Salt Lake City, UT), a manufacturer of telecommunications equipment. "They go to the store and pick some NAPA parts and some GM parts and some Ford parts, and they build their own custom car." As a result, the average system administrator has a nearly impossible task. You need a product that can, first, create a model of your patchwork n
etwork and, second, let you simulate what-ifs to that network. So what do you do?
You must zealously identify the tools that can grasp your network's complexities, then assemble those tools into a system that works. Large networks are built out of disparate hardware platforms, operating systems, and applications, and managing them is a daunting adventure. Without the right tools, it's downright impossible.
Where to Find
CenterLine Software
Cambridge, MA
Phone: (800) 669-2687 or (617) 498-3000
Fax: (617) 868-6655
Internet:
http://www.centerline.com
Building the Model Network
