Previous | Table of Contents | Next |
If the data center manager has past experience with free software bundled with other products, he or she is aware of the shortcomings. Anyone who has telephoned vendor help lines for popular PC products knows there may be a considerable wait before the call is even answered. Time, according to the old expression, is money.
Support costs are not trivial. According to a recent report on the costs of software ownership by Dataquest, the average annual support costs for networking and communication software for the Microsoft Windows environment are nearly twice as much as the initial product street price. This figure is high, compared to office suites, such as Microsoft Office, with support costs that are less than half than the initial product price.
The higher costs for networking support are easy to understand when the support policies offered by leading software firms are examined. One major vendor that includes a free stack with its operating system charges $35 per incident for networking technical support. This charge applies from day one; the vendor excludes networking questions from the 90-day free support policy it typically offers for other software.
One alternative is to purchase a support plan. Again, the same vendor offers a support plan for larger organizations at $25,000 a year, for up to 150 incidents a year, with an additional $1,500 for 10 additional incidents. This means that free stacks can be very expensive to support, especially in the long term. Data center managers should look for a company that includes free support past the 90-day warranty period and that offers online help and fax-on-demand services.
Networking software is very different from word processing or spreadsheet software. Those programs are unlikely to bring down the entire network if they fail. However, even with the best networking software, the network will inevitably go down on occasion. When that happens, data center managers need diagnostic tools to get the network up and running again quickly, and they need monitoring capabilities to avoid future problems. Free stacks typically lack desktop monitoring and diagnostic tools.
If the organization needs critical features such as diagnostic and monitoring capabilities, file sharing and/or IP addressing, a stack with these features must be purchased, because these features are written to the kernel, not to the WinSock API. Managers cannot simply plug in desktop management tools, NFS, or BOOTP from one vendor and use them on anothers stack. This means an industrial-strength TCP/IP tool is the solution.
When a TCP/IP stack is bundled with a personal operating system customers may have to wait for a new version of the operating system before enhancements are added to the TCP/IP software. Historically, that can mean long waits. Yet in the TCP/IP and Internet world, new specifications and applications are being added all the time.
Networking vendors should be able to respond more quickly, delivering upgrades, feature enhancements, and patches faster, so your corporate network will not suffer while awaiting the next release of an operating system.
Once these questions have been considered and answered, the organization must test the selected stack. There is no substitute for trying software in the actual environment. With sufficient evaluation, following the criteria outlined in this section, the TCP/IP stack will provide the reliability, scalability, and flexibility required by corporate networks for their mission-critical applications. While such high-value software may initially cost more, it is an investment that will save time and money in the long term.
Conventional wisdom says that all TCP/IP stacks are created equal. Whether they are embedded into operating systems, available free as shareware, or sold as part of a TCP/IP application suite, they are all interchangeable.
This view takes into account only that portion of the TCP/IP stack that has to do with Windows Sockets (WinSock). WinSock is an open API that functions like a standard wall outlet, giving developers the freedom to write TCP/IP-based applications and plug them in without having to worry about the peculiarities of the underlying TCP/IP stack. Common WinSock applications include Internet browsers, FTP clients, PC X servers, and VT420, TN3270, and TN5250 connectivity products. Written to the WinSock API, these applications should work with any TCP/IP stack.
WinSock is just the surface of the TCP/IP stack, however, and it is not the only portion of the stack to which applications are written. Below the WinSock layer, TCP/IP stacks are largely proprietary, with different architectural approaches to how they meet user needs. In addition, many applications that are important to the smooth functioning of an enterprise network are integrated into TCP/IP below WinSock. These kernel applications are the TCP/IP issues that are largely ignored, and the silence is costing companies both money and time.
Many applications can be written only to the kernel; network management tools are one example. This means that a vendors management tools will work only with its own stack. Therefore, a data center manager who is interested in being able to manage hundreds or thousands of desktops running TCP/IP will most likely want to select the stack that has the best network management applications built in. An organization cannot simply purchase and patch in applications to a free kernel. This is also the case with many other applications, including:
The number of network-critical applications that exist at the kernel level is substantial, and because an industry standard kernel interface does not exist, there is no way to pick and choose among kernel-based applications to optimize a stack. If the organization now, or in the future, plans to use any of these applications, data center managers need to carefully evaluate their TCP/IP stack decision.
Previous | Table of Contents | Next |