InformationWeek
Defining The Business Value Of Technology
November 15, 1999
 Printer ready |
Take Your LAN For A Test Drive
Ganymede's chariot measures the effects of network applications before you install them
| Related links: |
|
|
| And from our sister publications: |
|
|
lanning to make changes to your network? Perhaps you want to add a Web server or implement an enterprise resource planning solution, but you wonder how such an adjustment will affect your network's performance. How much throughput can your campus backbone carry? What kind of impact will an SAP server have on your Web servers? What will happen if you consolidate two file servers into one box? Would replacing your Fast Ethernet backbone with Gigabit Ethernet really make a difference? Can your existing infrastructure support streaming multimedia?There's no easy way to measure your network's capabilities or predict how well it will perform in the future--at least, not without simulating specific types of network traffic. Fortunately, Chariot 3.1 from Ganymede Software Inc. provides just the sort of sophisticated tool you need to accomplish this task.
Chariot 3.1 sends traffic between pairs of servers or workstations, then reports on system performance. Companies that get beyond gasping at Chariot's high initial cost will reap equally high returns--as long as they also have human support talent on hand.
Chariot is conceptually simple and deceptively powerful. The product consists of two software elements: Chariot Console, for setting up tests and viewing results, and Chariot Performance Endpoints, tiny bits of code that run on many types of servers and workstations.
To illustrate the way Endpoints and Consoles work together, install Endpoints on two PCs (specified by their names or IP addresses), then install Console on a third PC. From the Console, select a simple traffic test--for example, faux transactions between fictitious RealAudio servers and clients. When Chariot runs the test, the two PC Endpoints conduct the transactions and report the results to the Console, which, in turn, discloses bandwidth consumed, number of transactions conducted per second, load handled by each Endpoint processor, variability of packet delivery (or "jitter"), and other critical data.
If that were all Chariot could do, it would hardly justify its $14,000-per-console price tag (Endpoint licenses are free). Chariot's real value lies in its ability to accommodate dozens, or even hundreds, of Endpoint workstations and servers in a single test. Each Endpoint can run dozens of point-to-point transactions simultaneously, thus establishing a test bed for a number of real-life situations.
We installed Chariot Endpoint software on five Windows 98 and NT workstations, and on one Windows server in a shared Fast Ethernet network, an IBM Netfinity 5500 with dual 400-MHz Pentium II processors. We then configured a test in which each Endpoint workstation ran 10 simultaneous database queries against the Netfinity--in effect, simulating a total of 50 end-user workstations. We steadily increased the number of simultaneous end-user transactions until we reached a point at which the server couldn't handle any more load (in other words, adding more clients didn't increase the transaction rate). The practical benefit: We now know how much work we can get out of that server.
As a result, we not only determined the load level of the server, but also developed the means for testing different network topologies or comparing the performance of different hardware configurations. Would adding more RAM help? What about substituting a different CPU or adding a second Fast Ethernet connection to the local area network? Chariot let us test all those variables--assuming, of course, that we had more RAM, processors, and network interface cards available to create those scenarios. For additional realism, we could have generated additional network traffic, using other simultaneous Chariot Endpoint-to-Endpoint transactions.
Nearly every aspect of Chariot is configurable--which illustrates both its greatest strength and its greatest weakness. On the plus side, it's possible to design tests that can approximate virtually any software configuration implemented on the network's current physical infrastructure. But the associated disadvantage is that it's difficult to know exactly how to configure the test to reflect your real-world goals.
Unlike other design tools that simulate a network entirely in software, Chariot makes use of every element in the network infrastructure, from switches, routers, and servers to protocols and traffic. This makes it easy to test the impact of adding a new application to a LAN but difficult to run what-if scenarios that involve changing the LAN, such as testing the impact of an asynchronous transfer mode or Gigabit Ethernet backbone when one does not already exist. When the hardware is there, though, Chariot can stress it.
At the same time, Chariot's applications aren't limited to the LAN. As long as the Endpoints can talk to each other and to the Chariot Console, you can run test suites. Although we didn't test this feature, the documentation describes how to run performance tests through a firewall or across a wide area network or a virtual private network. Used this way, Chariot becomes a tool for monitoring service-level agreements or for testing how well Web servers deliver the goods.
Although Chariot Console requires a Windows 95/98 or Windows NT workstation, Endpoints can be run on nearly any platform--including all flavors of Windows, OS/2, NetWare, Solaris, HP-UX, AIX, and Linux; the only notable omission is Macintosh.
Every test transaction between a pair of Endpoints requires a test script that contains short software programs for each Endpoint. Chariot ships with a variety of predefined scripts for testing traffic analogous to client-server transactions, unidirectional broadcasts and multicasts, Lotus Notes and SAP R/3 sessions, simulated network logons, E-mail retrievals, database queries, file transfers, long sessions, jitter-sensitive streaming media--44 tests in all. The scripts are simply written to be easily understood by anyone familiar with Basic or Pascal. Ganymede also includes a built-in editor so you can edit or create scripts yourself. It takes only a minute, for example, to modify Chariot's HTTPGIF script to change file sizes and frequency.
The true limitation of Chariot's technology is human knowledge. If you're just trying to load up the network to see if it breaks, Chariot is a breeze to use. But if you're trying to create a prototype of a specific arrangement--for example, streaming multicast video deployment--the challenge lies in knowing which tests to run, how many instances of each test you need to write, and how to configure them. It can be done, but it's vital to invest sufficient time to understand how Chariot works and to familiarize yourself with the character of each of the test scripts.
The payoff for any test tool is evident in its reporting, and Chariot does not disappoint. The graphical user interface displays tests using line charts and tables. The charts show peaks and valleys that might indicate problems with servers or workstations, or with how network devices handle specific types of traffic. The information may also be displayed in tabular form to view or compare the total bandwidth of transactions at a specific Endpoint, for instance, or the speed of individual transactions. By running tests at different time intervals, or by rerunning them after changing the network configuration, Chariot gives network managers a tremendous analytical advantage.
It's unfortunate that Ganymede doesn't offer a lower-priced version of its testing software--say, a "Chariot Lite" that's limited to 50 simultaneous pairs in a test and is priced at $2,500. Such an expense would certainly be easier for enterprise managers to justify to their superiors.
For now, though, it's all or nothing. Still, if you're contemplating revamping your network or need an evaluation of its ability to support your future needs, Chariot's price tag might well be considered a bargain.
Alan Zeichick is principal analyst at independent technology research firm Camden Associates. He can be reached at zeichick@camdenassociates.com.
Back to Labs
Send Us Your Feedback
Top of the Page
