Thin-Client Computing, Coming Full Circle (Part 2)
NT continues to gain market acceptance as a host for mission-critical applications in the enterprise. As more success stories are published on thin-client problem-solving and TCO reduction, the chasm will be crossed to an era of broadly accepted thin-client and server-based computing.
This is part two of a two-part article on thin-client computing. Part one laid the foundation for thin-client computing, covering its history, the relationship of Microsoft and Citrix, and the basic pieces of the solution. What better way to begin part two than by starting with what gets all the headlines.
Although the usefulness of thin client is driven more by the capabilities of the server, a still-important element in the thin-client computing model is the client-side computing devices. In 1998, 507,000 thin-client devices were shipped, compared with over 71 million desktop PCs, which corresponds to just under 1 percent.
It is projected that in 2002, almost seven million thin-client terminals will ship, compared with around 115 million desktop PCs, representing almost 6 percent. This is an astonishing 82 percent compounded annual growth rate for thin clients. By 2002, the installed base for thin-client terminals will be 15 million.
A main defining characteristic of thin-client devices is how they operate without a network – that is, when the network is down or the user cannot conveniently connect to the network. Here are the prominent thin-client systems.
A "standard" Windows-based PC is your typical Windows "9x" PC that runs full applications, but has relatively high up-front hardware costs and high support costs. By continuing execution of the many existing local applications on the PC, the PC continues in its present role. Then, by running the client-side thin-client software on the system, access to thin-client server-based applications and databases is provided. The client software manages the input and output interactions. While traditional PCs can operate without a network, the applications available only over the network would not be available when the network is not available. Using older (dare we say obsolete?) PCs as thin clients extends their lives and maximizes the investment return.
A "locked-down" PC is just like a standard one, but the user is limited by software to the amount of customization that can be performed. Applications are available and executed on the server, which is where the related disk storage is.
A NetPC is a stripped-down (read that as reduced-cost) PC in the sense that only limited disk storage is provided and very few other options are provided, such as there are no floppy or CD ROM drives. NetPCs came about as a knee-jerk reaction by PC vendors to the threat posed by the network computer’s reduced price.
NetPCs do cost more than network computers. There are very few adherents to this technology now, though. Because NetPCs do have local storage, they can operate independently of a network.
A network computer (NC) breaks completely from the previously defined types of thin clients. Whereas, the preceding are truly PCs, the NC and what follow are not. They have no user-accessible disk and cannot run applications independently. Thus, they depend entirely on being connected to a network for data storage and application execution (i.e., to perform any useful work).
Although not as low cost as initially desired, NCs do generally cost less than the many varieties of PCs. NCs depend on network speeds for transmission of input and display information. They also lack the ability to be customized, as PCs can be.
NCs generally represent a reduced total cost of ownership (TCO) when compared to an equivalent architecture of networked PCs (refer to the TCO section later).
An NC is an intelligent device that runs a Java or similar environment. A Java NC, not surprisingly, is optimized for running Java applets. It provides cross-platform capabilities, tied to the Java language’s and Java Virtual Machine’s (JVM) capabilities.
When booting, the Java NC downloads applets for local use and connects over the network to the applications that are on the server. However, Java still lacks generally acknowledged support for mission-critical applications.
A Windows-based terminal (WBT) is naturally compatible with and optimized for Windows applications. A WBT has sometimes been referred to as a Citrix NC. The server it connects to is a Microsoft Windows NT server; in contrast, an NC is server neutral. The Unisys WinPath is an example of a typical Windows-based terminal. The WBT and NC are similar in their positioning as alternatives to the standard PC.
The most recent addition to the thin-client terminal space is the ultra-thin client. It is not yet widely accepted, but is an interesting variation on the thin-client theme.
Essentially, the ultra-thin client is a low-cost display, mouse and keyboard connected directly to a card in the server, which performs all of the required processing. This is a complete return to the pure dumb terminals of mainframe land.
The terminal market is characterized by high competition between suppliers, market consolidation, and increased commoditization of technology. Thin clients are expected to be the terminal of choice in corporations, compared with other terminals, such as 3270-style and asynchronous (ASCII/ANSI-based) character terminals. And the largest percentage of thin-client shipments commencing in 1999 will be Windows-based terminals.
For example, Unisys is a user of thin-client technology. Unisys was faced with rolling out PeopleSoft human resource application access to 1,700 of its managers. The problem facing Unisys was an aging desktop population (speed, disk size and memory), application performance on the desktop, and the cost of rolling out and managing the application.
By selecting a thin-client solution in conjunction with PeopleSoft, Unisys saved $2.5 million in PC and other hardware upgrade and administration costs. Furthermore, the typical LAN-connected manager experienced a 200 percent to 600 percent performance improvement compared with running the PeopleSoft application locally on the desktop. This eliminated the poor performance associated with heavy client-to-server network traffic.
Perhaps, best of all, the Unisys PeopleSoft project schedule was dramatically reduced compared with a typical client/server project. This was because the thin-client implementation required no manual upgrade to take place at the 1,700 desktop locations.
It has been widely reported that Federal Express has long been a proponent of thin client. FedEx has installed six thousand Windows terminals in airport ramps, hubs, and city stations, and plans to deploy up to 40,000 units within the next few years. They are moving toward networked solutions that also allow for employee mobility.
In addition, FedEx chose Windows terminals since Windows is a familiar environment for the employees. FedEx plans to continue to use the existing tens of thousands of PCs and laptops for their various professional functions.
Total Cost of Ownership
TCO definitely favors the network computers from a desktop perspective. In December 1997, the GartnerGroup came out with what is generally considered a landmark thin-client TCO study. (It was updated in late 1998.) One of its goals, which it satisfied admirably, was to show that there was more to desktop systems than just the purchase price of the system. The study was definitely focused on a moving target, though, considering the fluidity of pricing and constant new product introductions.
The Gartner study took the following areas into account: desktop capital, technical support, administration, end user operations, network capital, network technical support, network administration, and network end user operations. These areas covered the annual costs to implement, upgrade, maintain and support the desktop system. The results were startling because some areas increased in cost for thin-client systems, that might otherwise have been overlooked.
In general, a thin-client implementation saves somewhere between one-quarter and one-third over a Windows 9x distributed computing model. Even factoring in the inclusion of the required servers, thin-client computing is more cost-effective on a TCO basis.
It was not an across-the-board win in all categories for thin client. While desktop capital, technical support, administration, end user operations, and network end-user operations were all lower when a thin desktop was selected, network capital and network technical support were both higher.
The TCO methodology can be applied to an existing PC being used as a thin client. In this case, while still broadly favoring thin client over a client/server rollout, the numbers marginally favored using the existing PC over a new-NC implementation.
Be aware that there are significant differences in the way the WinFrame product is licensed versus how the combined Terminal Server and MetaFrame product set is licensed. This affects the cost structure and TCO, reducing the per-desktop TCO slightly and perhaps influencing which to select.
The reasons supporting the reduced thin-client paradigm costs include:
• Thin-client desktop systems generally cost less than fully capable PCs – although the desktop capital reduction is only a portion of the entire reduction.
• Thin-client terminals have a lessened threat of obsolescence, since they have no storage, memory, etc.
• Thin-client desktops have reduced technical support demands since a terminal is simpler (i.e., it has less capability), and, therefore, generates fewer issues, questions and problems.
• Thin-client desktops are easier to administer and manage, being largely centralized. Reductions of up to 80 percent can be experienced, which includes a corresponding reduction in the required support staff. For example, file backup is centralized, and there is no time-consuming application installation required on multiple desktops during a rollout, since the application is installed only once on the server.
• Desktop maintenance costs can be controlled. If a thin-client system fails, it is usually replaced in its entirety, and the user is up and running in only minutes, connecting to the server where all the applications, databases and files reside. This is compared with a standard PC system that often requires a complete and time-consuming rebuild of the hard drive, including installing and configuring all of the applications and customizing them to the user’s requirements and wishes.
• System upgrades, including the associated capital expenditures and installation time, are minimal for thin-client systems.
• Operating system configuration, performance tuning, and maintenance are easier for thin clients.
• Auditing is easier, since there is no software or removable parts to be inventoried on the thin client.
• Sensitive data can be at risk when residing on numerous uncontrolled PCs. With a thin-client implementation, critical files are stored and backed up on the server, making them more secure.
• Security is also more manageable with thin clients, since, other than being an access portal, there is nothing of value (i.e., no data) stored on the thin-client system. Centrally located servers can be made physically more secure than dispersed desktop systems.
• Thin-client hardware is a potentially less-interesting target for an equipment thief.
• Virus protection is a non-issue since there are no files stored on a thin-client system. It is controlled at a central point, on the server.
• System documentation and end user formal training and casual learning is simplified.
On the complete flip side, there are additional requirements to support a thin-client architecture. For example, highly resilient and highly available servers are required. And the network to which the thin-client system connects must be more robust, since all desktop work now depends on the network.
These additional requirements drive related requirements. For example, additional server storage, such as that provided by a network-attached storage (NAS) device, and other network hardware, software and associated services may be required. There may also be a corresponding decrease in, for example, desktop storage needs.
To what environments is a thin-client computing model best suited? Broadly applicable examples include new applications for character-based terminals (terminal replacement), extending a PC’s useful service life (investment protection), front-ending mainframe applications, improving an application’s network performance and assisting in a more-rapid application deployment.
Users with legacy "green screen" terminals, such as 3270, UTS, or T27, do not have access to the latest Windows applications. Nor do they have e-mail capabilities or Internet or intranet access. Yet many of them are now using Windows PCs at home and desire that familiar GUI environment in the workplace.
In this case, the terminals are replaced with WBTs that provide access to the legacy applications as well as the new Windows- and browser-based applications. This can be done as described in the total cost of ownership discussion for far less than a "fat" PC deployment.
Many organizations are faced with an enormous number of older PCs of the 486 or even 386 variety. Perhaps, there is even a mixture of DOS, Macintosh, and Unix desktop systems. The problem here is that these functioning though aging systems cannot run current software (such as Office 2000 or Outlook) without a significant upgrade or complete replacement.
To protect this huge investment and to provide cross platform access to Windows applications, these old systems connect to a thin-client server running the desired software suites. From an end user perspective, the server-based application appears as if it is running on the desktop. Unisys used this strategy, as described in the examples.
Thin-client servers can be used as a front-ending device that connects Windows-based terminals seamlessly to an existing mainframe application. For example, Attachmate terminal emulators allow the thin-client desktops to appear to a ClearPath system exactly as before, except the users can now also use the latest Windows applications without requiring a PC on every desk.
By using a ClearPath high speed channel adapter, the connection between the thin-client server and the ClearPath is very fast. Additionally, the Unisys MAPPER Presentation Client and PowerClient runtime products can be run on the thin-client server with some restrictions.
When a client/server application is run over a network, there is a considerable demand placed on the network and its bandwidth, as database records, files, and even entire databases are transferred across the network. The same applies when remote (mobile) users access the application. This causes long response times. By using a thin-client architecture, only the input and display information is transmitted.
This results in an often-sizeable performance improvement that not only can prolong the life of the existing network, but also provide the employee and therefore an entire organization with a significant increase in productivity.
If you have ever participated in the rollout of a new client/server application to a distributed, network-connected, and geographically dispersed (multi-site, multi-national) set of desktop systems, then you are familiar with how long and how difficult such a deployment can be. It is hard and costly to manage applications on distributed desktops. This includes managing, such tasks as physically installing the application on each desktop, testing its connection to the database, and, worse, coping with the myriad of desktop types, operating system versions, and unique and unusual (non-standard) desktop customizations and configurations.
With a thin-client deployment, the application is installed once on the server and, logically, a switch is turned on for each desktop being given access to the application. A single location for the application to reside makes testing, backup and recovery easier. It also reduces the complexity of ensuring software license compliance.
To the end users, a thin-client deployment means always having the most-current application version, and it means eliminating the updates to the desktops every time a revision to the application is received. It reduces, for example, the difficulties in sharing documents caused by multiple incompatible versions of the word processor, since everyone receives the new word processor at the same instant.
The result is that applications can be rolled out more quickly and confidently, providing a competitive edge to the organization in this era of "Web time."
In addition, certain vertical markets seem better suited to a thin-client architecture, including finance, public administration, and community services. These segments share similar characteristics, including many people repetitively performing similar and structured tasks.
This implies a limited need for a flexible and customized desktop system. Other peripheral areas for considering thin client include repetitive data entry, education, and small-to-medium sized businesses.
Users Fall Into...
When considering the rollout of a thin-client architecture design, the IT manager must think of users as falling into one of three categories. This helps determine many configuration factors associated with the rollout.
A "light" user (task oriented) is similar to a call-center-application user. Here, a script leads the user through a sequence of questions that accesses and updates an associated database. Perhaps, printing is involved.
A "medium" user (administrative) is one that creates small documents, enters data in a spreadsheet, and sends and receives a moderate number of e-mails. There is minimal Internet and intranet browsing.
A "heavy" user (knowledge) is one that has multiple applications open at the same time, creating and editing several documents, spreadsheets, and presentations simultaneously, and does graphing and charting of results.
A thin-client design must take into account many factors and situations. Here are a few of them:
• Number of users to be supported
• The number of servers required, one to run the Terminal Server software, and others, as required, for primary and backup domain controllers; multiple Terminal Server hosts can also be used for both redundancy, as well as for load balancing
• Considerations for applications targeted for the server; this is multifaceted – for example, ramifications of 8- and 16-bit application support
• Which applications are out-of-the-box ready to run in a thin-client environment, such as PeopleSoft
• The intended primary mode of use: as an application server, remote access server (RAS), Internet gateway, front end for another NOS, etc.
• Interaction with applications resident on the desktop
• Database sizing
• Interoperability with the existing network operating systems (NOSs)
• Directory services/domain structure
• Devices that will be accessing the thin-client server (vintage desktops, NCs, WinPath terminals, etc.), determining the standard desktop configurations intended for server interaction
• Network system hardware configurations by location so the infrastructure can be integrated seamlessly with the thin-client architecture
• Communications infrastructure
• Inventory of current network components, protocols, and services
• Site readiness and requirements
• Test everything thoroughly before rolling it out to the users.
About the Author: Charlie Young is Director of U.S. Network Enable solution programs in the Global Customer Services (GCS) organization for Unisys. He can be reached at email@example.com.
Starting a Thin-Client Project
By Charlie Young
Clearly, a pilot is the best method for starting a thin-client project. Make sure you understand the total cost of ownership issues and the strengths and weaknesses of all the technologies.
A complete solution will consist of the following:
• One or more preloaded and tested, properly sized, enterprise-class (fault-tolerant and scalable) servers
• Servers configured and tuned specifically for thin-client applications with high up-time guarantees – thin-client servers need to be very reliable, given the critical nature of what they’re running and who is depending on them
• Comprehensive server maintenance
• Uninterruptible power supplies (UPS)
• Microsoft Windows NT Server Terminal Server Edition software, along with necessary NT desktop licenses
• Citrix MetaFrame software
• Network changes, including hardware and wiring changes
• Thin-client terminals, such as Windows-based terminals
• Software for load-balancing between thin-client servers
• Software for system fault tolerance and NIC (network interface card)
• Attachmate emulators
• Administrator training
• Clear and unbiased end user education used to explain the differences between standard PC desktop computing and thin-client computing; quirks must be emphasized
• Appropriate consulting for network assessment and design (LAN traffic patterns change as a result of implementing thin client), LAN and WAN integration, thin-client architecture design, determining appropriate and inappropriate applications, application adaptation, partitioning and tuning, application testing and potential application scripting, server and network capacity planning and tuning, high-availability server designs and performance requirements, and NT migration
• Consider implementation services and project management for large roll-outs
• Outsourcing, to run the entire operation or just a piece, such as network or help desk management
• A vendor that can supply a complete hardware and software solution, including related services, not just one of the many pieces required for a complete solution
And For More Information
For more information on Microsoft Terminal Server, check out: www.microsoft.com/ntserver, and navigate to Terminal Server Edition. For application compatibility testing, try www.veritest.com/terminalserver. For clients for non-Windows platforms, visit www.citrix.com. For terminals, go to www.wyse.com.
Remember to focus on the business problem that needs solving, not on the hype created by the thin-client desktop manufacturers. And do not pay attention to their failures, sometimes spectacular, often covered broadly by the trade journals. Thin-client computing provides a serious solution for serious problems and offers potentially huge savings.
As to whether thin-client computing is a continuous or discontinuous innovation, the market ultimately decides. On the one hand, the mainframe computing model – nowadays known as thin-client or server-based computing – has been around just about forever. But, frankly, it is a new computing paradigm for Microsoft Windows NT. Just as minicomputers did not replace all mainframes and PC-based LANs did not replace minicomputers, thin-client computing will find its place among them, not replacing them.
To emphasize its growing acceptance, network-centric applications are already being developed by mainstream organizations that enable the thin-client computing paradigm. Additionally, host-based and client/server applications are becoming network-enabled through HTML, Java, ActiveX, and the like. And pure network-centric applications, that are location-independent and integrate relational databases with Web sites, e-mail capabilities and telephony, are just now being developed by the early adopters.
NT continues to gain market acceptance as a host for mission-critical applications in the enterprise. As more and more success stories continue to be published that clearly articulate how thin client solved this or that problem, reduced TCO, or improved desktop management, the chasm will be crossed to an era of broadly accepted thin-client and server-based computing.