The client/server world measures a client's fatness by how much of an application runs on the workstation. According to industry analysts, fat increases total cost of ownership (TCO) because of the maintenance required to keep the client workstation running. Oracle and Sun Microsystems use this analysis to claim that network computers (NCs) reduce the fat, and therefore, can save organizations money.
CUT THE FAT
TCO has sounded the biggest wake-up call to the PC industry since the Internet. For the past year, the PC industry has been reacting to the TCO message. TCO has fueled the fire behind NCs, Network PCs (NetPCs), managed PCs, and more. Although PCs have increased user productivity compared with dumb terminals, PCs have also increased systems management costs. In a recent study, "The Real Cost of Network Computing" (Computerworld and IDG Research), most IS managers said that implementing PC-based networks has increased costs for staffing, hardware, software, training, support, consultants, and network maintenance. In response to this situation, TCO awareness has pointed a finger at suppliers and said, "Enough."
TCO causes businesses to rethink how to deploy clients in the enterprise. You face several strategies, which Figure 1, depicts. I group client strategies into three categories: thin, lean, and fat. I'll introduce these strategies, present key players in the industry, and focus on thin clients. A future article will concentrate on lean and fat clients, and how to decide which strategy is best for your environment.
Buying a client device is like buying a container of milk: You must decide whether you want skim, 2 percent, or whole. Regardless of the solution you choose, it must meet certain goals to reach TCO objectives: Client devices must be truly network-centric, letting anyone access anything, at anytime, from anywhere; users must be able to access legacy applications safely and securely from any location; you must be able to centrally control desktops to allow for easier software maintenance and upgrades; your solution must be secure and, of course, reduce costs.
A thin client is a network-dependent terminal capable of displaying remote applications running on an attached server. All applications, processing power, and user configurations reside on the server.
Imagine rolling out an Office 97 upgrade to your enterprise in only two hours. This scenario is the promise of thin client technology because you register the application with only a central server. You set all your user profiles and configurations at the server. Office 97 runs on the server and sends only screen updates across the wire to the terminals. You change nothing on the client devices.
Contrast that scenario with a typical PC client-based enterprise network. Even if you run Microsoft's Systems Management Server (SMS), each user must respond to an upgrade notice, wait until the application installs on the local hard disk, and then answer the appropriate prompts. If you're not running a software distribution program such as SMS, you must install the upgrade at each PC. If the PC needs more memory, more disk space, or a faster CPU, your upgrade could take months.
Because a thin client doesn't perform any processing, thin client environments require huge servers. For typical business applications such as Office 97, a 4-way Pentium Pro-based server with 512MB of RAM can service 60 users (15 users per CPU and 128MB of RAM). In a fat client environment, that 4-way server could handle 600 file-and-print users.
Stuck on NT Server 3.51
Today's thin clients are based on multiuser Windows NT Server extensions from Citrix Systems and Prologue Software. When you install these solutions, you install a modified version of NT Server 3.51 that includes the multiuser application server software as services in the core OS. Prologue replaces 29 files, and Citrix replaces 430 files from the standard version of NT Server 3.51. Both solutions modify NT's kernel (ntoskrnl.exe).
When Microsoft released NT 4.0, both Citrix and Prologue sought to build their solutions on this new version. Microsoft no longer wanted modified versions of NT in existence, seeking to have total control of the operating system. For a while, how Citrix and Prologue were going to update their NT Server 3.51 multiuser extensions was unclear. Last spring, Microsoft licensed Prologue's software and Citrix's multiuser extensions. This licensed code forms the basis for Microsoft's thin client solution (code-named Hydra), which I'll describe later in this article. As part of the license, Microsoft has agreed to upgrade existing Citrix and Prologue sites with Hydra for NT Server 4.0 (and future versions of NT).
Unfortunately, until Hydra ships (some time in 1998--beta 1 will come out in the fourth quarter of 1997), the only thin client solutions available are based on NT Server 3.51. Therefore, any application that requires NT 4.0 on the client won't work.
Citrix has created an NT-based thin client market by shipping over 30,000 WinFrame servers. In a recent survey of Windows NT Magazine readers, 6 percent of you stated that you have Citrix-based servers, and an additional 6 percent of you plan to purchase Citrix-based servers over the next 12 months.
The current Citrix software includes three parts: NT Server, multiuser NT extensions (MultiWin), and Intelligent Console Architecture (ICA) client/server technology. For NT Server 4.0 and later, Citrix-based solutions will include NT Server, Hydra, and ICA. Citrix retained the rights to ICA, a core client/server technology that supports a large number of devices. The ICA client, which is available for NT, Windows 3.11, Windows 95, DOS, Macintosh, and UNIX, is a small piece of software that lets client devices display information sent by the ICA server. The ICA protocol sends only keystrokes and screen updates across the wire.
Why didn't Microsoft purchase the ICA client/server technology? "ICA is Citrix. To get ICA, Microsoft would have had to buy Citrix," said Mark Templeton, Citrix's marketing vice president. ICA clients can run anything that can run on NT Server 3.51, including Java-based applications.
Citrix has licensed its technology to several terminal manufacturers who have embedded an ICA client into their hardware devices. For example, Wyse Technology has created an ICA-based terminal called Winterm. Early adopters who bought Winterm clients when 486s were the standard have already bypassed two complete upgrade cycles. They simply upgraded their WinFrame servers and application software. They didn't need to make any changes to their Winterm clients.
Citrix has also licensed its WinFrame server technology to Tektronix, Network Computing Devices (NCD), and Insignia Solutions. These companies have added capabilities to WinFrame to emphasize specific areas of the market, including non-Windows clients such as Macintoshes and X-terminals.
NCD adds X-Windows support and other features to WinFrame and calls its server product WinCenter Pro. With the NCD solution, you can use existing X-terminals, commonly found in UNIX shops. WinCenter Pro lets thin client users copy text between X-Windows and Windows applications. NCD also manufactures X-terminals, NCs, and Windows-based terminals. When coupled with WinCenter Pro, this hardware and software combination provides a thin client solution.
Like NCD, Tektronix adds X-Windows support to WinFrame and calls it WinDD. WinDD has created ICA clients for the seven major UNIX platforms. Tektronix manufactures a variety of thin clients with protocol combinations. For example, you can buy terminals that natively support a combination of X, ICA, and 5250. With this combination, one terminal can attach to a UNIX Server, a WinDD server, and an AS/400, and display applications from their native environments.
Insignia adds support for Macintosh, X-Windows, and Java-based NCs to WinFrame and calls its product NTRIGUE. The goal of NTRIGUE is to let any device (fat, lean, or thin) run Windows-based applications. Insignia's newest product, Keoke, is a Java-based X-Windows terminal emulator that lets NCs and Java-enabled browsers access Windows applications via NTRIGUE. "We prefer to be the arms merchant in the battle between NCs, Windows terminals, and PCs. We support them all," said Peter Crosby, the product manager for Insignia's NTRIGUE.
Application configuration is a big concern for multiuser application servers. For example, software companies typically have not designed 16-bit applications to work in a shared environment. During installation, 16-bit applications assume they have total control of the PC and attempt to install .ini files in the system directory instead of a user directory. Similarly, some 32-bit applications write temporary files using fixed names in common locations (e.g., the TEMP directory), or they store information in HKEY_LOCAL_MACHINE instead of HKEY_CURRENT_USER.
Citrix intercepts about 80 percent of such installation problems, but the remaining 20 percent of the problems require workarounds. Insignia has dedicated a Web page (http://www.insignia.com/NTRIGUE/Application.html) to application configuration notes to help you overcome these problems. In addition, the company has developed an application configuration tool (ACT) to automate installing and configuring problem applications.
Prologue Software and Exodus Technology have co-developed the WiNTimes multiuser NT technology, which runs as an NT service on top of a modified version of NT Server 3.51. Exodus' NTERPRISE combines NT Server 3.51, WiNTimes, distributed display technology, and a client/server protocol.
Unlike Citrix, which uses the ICA protocol, Exodus bases its client/server protocol on X (more than 16 million terminals use X today). If a client can use X, that client can access Windows applications via NTERPRISE. In addition, Exodus supports the Alpha platform, which Citrix does not support.
Because of its distributed display technology, NTERPRISE can sense the native GUI of the attached client. For example, if an X-terminal is accessing a Win95 application, NTERPRISE configures the application to emulate the Motif window behavior: minimize, maximize, and close. This feature lets users maintain the look and feel they're accustomed to while using the application. Users can turn the multiwindow mode feature on and off from the client.
Another advantage of NTERPRISE is Exodus' pricing model. Currently, Citrix-based solutions charge a client access license (CAL) fee for every session. In contrast, Exodus charges a CAL fee for every client, regardless of the number of sessions that client uses. For example, 10 users, each of whom logs on to the multiuser server twice, generate 20 logons, or sessions. Citrix charges you for 20 CALs, but Exodus charges you for only 10 CALs.
NTERPRISE installs over an off-the-shelf version of NT 3.51 and modifies NT during installation. In contrast, Citrix bundles NT Server 3.51 with its product. Therefore, when you compare total solution prices, you must add the cost of NT Server 3.51 to the price of NTERPRISE to compare the products equitably.
Exodus positions NTERPRISE as the preferred platform for running networked, non-Windows clients, such as X-terminals. In performance tests, several vendors found that the X11 protocol performed better for locally attached thin client devices. ICA, with its compression scheme, performed better in low bandwidth situations such as WAN, dial-up, Internet, and wireless environments. With its reliance on the X protocol, Exodus positions NTERPRISE for only locally attached devices. In the future, Exodus plans to add support for low bandwidth environments.
IGC and Maxspeed
Intelligent Graphics Corporation's (IGC's) MultiNode for Windows NT uses the WiNTimes engine to support terminals that Maxspeed manufactures. Instead of using ICA or X, Maxspeed uses a direct video connection at 32Mbps. An NT workstation with IGC MultiNode can simultaneously behave as a client to a main server, and as a server to Maxspeed's MaxStation clients. This technology eliminates network traffic overhead to the MaxStations. The NT workstation remains fully operable as a local workstation when you need to access a floppy or a CD-ROM.
Figure 2 shows a typical Maxspeed configuration, in which one NT workstation hosts multiple terminals via the ISA or PCI bus. Each slot can host a MaxStation connection card, which comes in 1-, 2-, or 4-port configurations. A MaxRack lets one slot expand into a 7-port MaxStation connection. Each MaxStation connects over CAT5 wiring to an RJ45 connector. You can place the MaxStation up to 500 feet from its host.
Maxspeed positions its solution as a low-cost point-of-sale (POS) application. For example, the total price of a four-terminal solution is less than $5000. Because IGC MultiNode for Windows NT is tied to the Prologue multiuser engine, this solution runs on only NT 3.51.
Microsoft bases its multiuser application server on the technology it licensed from Citrix and Prologue. Microsoft code named this project Windows NT Hydra, after the mythical multiheaded serpent. Analogously to the serpent, Hydra Server will support multiple Windows sessions from a common server. Hydra clients can be Windows-based PCs running Hydra client software, non-Windows clients (e.g., Mac or UNIX workstations) running ICA-based client software, or Windows-based terminals that contain a thin operating system--only enough to boot from the server and run either the ICA or T.Share protocol. (To run ICA clients, you must buy an ICA add-on from Citrix.)
T.Share is Microsoft's version of the International Telecommunications Union (ITU) T.120 conferencing protocol. Currently, the only application using T.Share is Microsoft's NetMeeting application. Microsoft believes that T.Share can be optimized for both local and remote performance situations, thus providing the benefits of both X and ICA.
Microsoft is creating Hydra for several reasons. First, Hydra lets Microsoft beat NC proponents at their own game. Because NCs require that Java-based applications run on the client, Microsoft can now label NCs as fat compared with Windows-based terminals that do not run software on the client.
Second, Microsoft stands to make another billion dollars on thin-client technology. More than 35 million green-screenterminals in use today could be converted to Windows-based terminals. Applications such as hotel registration, POS, order entry, and warehousing are excellent candidates for these devices. Some vendors estimate that 20 percent of all desktops will be thin clients, which would make Hydra and Windows-based terminals a huge growth opportunity for Microsoft. Although traditional PC vendors may scoff at Windows-based terminals, such companies as Wyse, NCD, Tektronix, Citrix, and Insignia will make millions from Windows-based terminals.
Third, Hydra lets Microsoft reach its goal of Windows Everywhere. Today, you can run the latest 32-bit applications from Windows-based terminals, X-Windows terminals, Macintoshes, DOS-based systems, PCs, and more. You don't need to throw out your old 386s; you can use them to run 32-bit applications from the server.
Finally, NCs introduce an entirely new operating system: JavaOS. Because Java is both a language and an operating system, Microsoft can declare total support for the Java language, while declaring war on JavaOS. (In other words, Java applications are great as long as they run on a Microsoft operating system.) If JavaOS and NCs gained a solid foothold in the market, they could severely disrupt the Windows Everywhere critical mass.
Microsoft has licensed both Prologue and Citrix technology. Hydra will retain the Citrix user core, but both Citrix and Prologue developers are working with Microsoft on Hydra development. On one hand, Prologue works with few changes to NT 3.51. This factor has to be significant because Microsoft is trying to get its product out the door by the end of 1997. On the other hand, Citrix-based solutions have six years of history behind them and enjoy the market leader position. I believe Hydra's first release will be very close to Citrix, with additional support for T.Share Windows-based terminals. At press time, the specification for Windows-based terminals does not exist.
How does the T.Share protocol perform compared with ICA, X, or direct video? T.Share is still a relatively new protocol, but ICA and X have been optimized and have proved themselves for several years. Currently, T.Share is a Windows-only technology; however, software companies such as Insignia will be looking for opportunities to port the protocol to non-Windows environments (such opportunities might affect ICA over time). As part of the Citrix-Microsoft agreement, both companies have agreed to promote each other's technology. However, Microsoft will promote ICA only for non-Windows clients; for Windows-based PCs, Microsoft's strategy is clearly T.Share.
How will Microsoft package and price Hydra? Some industry analysts speculate that Microsoft will treat Hydra the same as Wolfpack and make it part of NT Server Enterprise Edition. This positioning would add value to NT Server Enterprise Edition and further distinguish it from the standard version of NT Server 4.0. Unfortunately, such positioning might make UNIX a less expensive platform for running POS applications with fewer than 25 users, the minimum level for NT Server, Enterprise Edition. My choice would be for Microsoft to make Hydra available for every version of NT, including NT Workstation, to provide a powerful, low-cost solution for POS, retail, reservation, and other task-oriented applications. Microsoft says this scenario is unlikely. After the first release, Microsoft will have a small-/medium-business version, but Hydra is focused at the corporate environment out of the gate.
Will Hydra price other solutions out of the market? Hydra will add two more CALs to any multiuser installation. For example, a future Citrix solution would include the price of NT Server, Hydra CALs, Citrix Server, ICA CALS, and NT Server CALS if you want access to NT Server file and print services. Regardless of how vendors price Hydra-based solutions, the total cost must be less than a competitive NC solution. Otherwise, Microsoft will lose its prime objective: to blow NCs out of the water--I mean, to reduce your TCO.
How will Hydra affect application developers? Application software vendors need to ensure that their applications behave well in a multiuser environment. The last thing they want is for Microsoft to point a finger at an application and label it incompatible. To help in this effort, Microsoft will publish development guidelines at the PDC this month and post them on its Web site.
How does Hydra support legacy applications? The combination of Hydra and Microsoft SNA Server would provide a compelling replacement strategy for 5250/3270 green-screen terminals. Combined with a Windows-based terminal, this solution would give users native access to IBM host-based legacy applications and the latest 32-bit Windows software.
Will traditional PC vendors create Windows-based terminals? At first, vendors such as Dell Computer and Compaq Computer will adopt a wait-and-see attitude. If they start losing customers, they will OEM units from existing vendors until the quantities warrant their own development. Creating NetPCs was easy for these vendors because NetPCs are still PCs. But creating a terminal is a whole new ball game. Clearly, Hydra offers the most opportunity to existing terminal vendors. You can be sure that Wyse, NCD, and other terminal vendors will have a Windows-based T.Share terminal ready by the time Hydra ships, in addition to their ICA-based terminals.
Buy Now or Wait?
Several reasons justify not waiting for Hydra if you want a multiuser solution. First, a majority of Windows NT Magazine readers support multiple platforms, the most common being UNIX. Because Hydra (out of the box) will not support terminals, Microsoft's Hydra presentations, recommend that you replace all your X-terminals with Windows-based terminals; but that's not necessary. If you don't mind the NT 3.51 user interface, X-terminals can run Windows applications. Next, all the current thin-client vendors will migrate their customers to Hydra, so even if you start with another solution, you'll eventually get to Hydra. As part of the Microsoft-Citrix agreement, all Windows-based terminals will be able to run the T.Share and ICA protocols. When Microsoft releases Hydra, T.Share will have fewer supported devices than X and ICA. If you don't need thin clients today and you want to attach only Windows-based terminals or older PCs, then wait.
A lean client is a network-dependent device that runs applications on the client and the server. This category includes NetPCs and NCs. Unlike a thin client, a lean client runs applications on the client device and can take advantage of the CPU power of the client. Because the client's CPU is critical to application performance, lean clients potentially face hardware upgrade cycles as applications grow and CPUs get faster. What user wants to be stuck with the slowest device?
However, a lean client is easier to maintain than a PC. As with a thin client solution, you get central control of applications, profiles, and security. In addition, because lean clients share the processing load, your server requirements are significantly lower than in a thin client environment. For example, a 4-way Pentium Pro-based server might serve 300 business application users instead of 60 thin client users.
The NC is a device that loads and runs Java applications from an attached server. The NC Reference Profile defines an NC as a device that has a minimum screenresolution of 640 * 480 pixels (video graphics adapter--VGA) or equivalent, a pointing device, text input capability, and audio output. In addition, an NC must support IP, HTTP, HTML, SMTP/POP3, Java Virtual Machine (JVM), and common multimedia formats (JPEG, GIF, WAV, AU). All NCs have a CPU, memory, and a way to connect to the network. NCs don't necessarily have persistent local storage.
The NC Reference Profile doesn't specify an OS, leaving that decision to hardware vendors. The two most common configurations are the JavaOS and the combination of NC OS and JVM. JavaSoft's JavaOS (http://www.javasoft.com) provides some consistency among NC implementations. Theoretically, applications running on two different JavaOS-based NCs will work identically. The JavaOS has a built-in JVM, the runtime environment required to execute Java applets.
The NC OS and JVM combination relies on the hardware vendor to come up with the OS and to implement a JVM for that OS. The Pure Java initiative has made progress in standardizing JVM implementations, but JVMs are inconsistent: The same Java application can behave differently on different JVMs--even with the same version number. The "write once, run everywhere" promise of Java isn't here today.
The advantage of the NC client model is potentially lower initial hardware costs. I say potentially because many NC vendors are adding multimedia hardware (e.g., sound, CD-ROMs, and video) to the device, raising the cost of an NC to that of a PC. The savings with NCs come from reduced maintenance costs and, consequently, reduced TCO. Most client system software downloads when the terminal boots, so with an NC you can centrally control and easily manage client applications.
A major disadvantage to the NC computing option is that applications must be written in Java. You can't run Windows applications natively on an NC. You need to add an NT-based multiuser application server solution. For example, IBM has partnered with NCD to manufacture NCs. With NCD's WinCenter Pro, IBM's NCs can provide connectivity to many environments, including 3270, 5250, Java, DOS, and Windows. Many NC proponents view this limitation as a temporary problem that will exist only until more Java applications are available. These proponents preach Java Everywhere with the same intensity that Microsoft preaches Windows Everywhere.
In a recent survey of this magazine's readers, 65 percent stated that you're planning to install thin or lean clients to run Windows-based applications. In contrast, only 13 percent of you plan to run Java-based applications. Currently, Windows-based applications outnumber Java-based applications 1000 to 1. This fact makes the NC Java-only strategy vulnerable, and is why thin clients, which can run both Java and Windows applications, are a compelling alternative.
In the future, NCs may become mobile. The Mobile Network Computer Reference Specification (MNCRS) provides guidelines for building mobile NCs, for use in situations without a persistent connection to the network. Unlike the standard NC, the mobile NC requires the instant-on feature, remote communications support, disconnected or local mode, and local cache storage (disk or RAM that lets the device work while disconnected from the network).
Microsoft's first response to NCs was to push the PC industry into incorporating some of the best ideas in manageability and TCO into a new PC design. This response put the PC on a quick-fix diet and resulted in the NetPC. A NetPC is an Intel- or Alpha-based PC that runs NT or Win9x and can remotely boot from the server. A NetPC requires PCI slots (no ISA), 16MB of RAM, an integrated network interface card, and the ability to lock the configuration (to prevent end-user changes).
However, to achieve its full potential, the NetPC requires a bunch of systems management software and firmware that will be implemented over the next 12 months. Until these features are available, you won't realize full TCO reduction from the NetPC. Dell has defined a phased strategy for implementing management features in its PCs, which will include the following features (I'll detail them in part 2 of this article):
* Phase I (today): Desktop Management Interface (DMI) 2.0 and support for Microsoft Zero Administration Kit (ZAK)
* Phase II (fourth quarter 1997): support for Intel's Wired for Management (WFM) Baseline Specification, which includes Wake-on-LAN; Advanced Configuration and Power Interface (ACPI); remote system boot and configuration; remote FLASH and BIOS installation; administrator password capability; and remote configuration of system drives, ports, and buses
* Phase III (May 1998, the same time as NT 5.0's release): support for Microsoft's Zero Administration Windows (ZAW), Microsoft's Windows Management Instrumentation (WMI), and the emerging Web-Based Enterprise Management (WBEM) standards
If venders incorporate these technologies into both the NetPC and managed fat clients, there's no advantage to a NetPC (and no difference between it and a managed fat client). The NetPC's value is that management is a requirement. To complete the package, however, you will need systems client management software and systems or network management software on the back end. (Part 2 of this article looks at management issues.)
New Moon's Liftoff
New Moon's Liftoff lets standard NT and Win95 desktops access server-based applications. Liftoff intercepts Windows API calls on the server and redirects the user interface (UI) and graphics device interface (GDI) calls to the client for execution on the client's Windows OS. Computational logic calls run on the server.
Liftoff is close to the Citrix model, and I would classify it as a thin-client solution except it requires a Windows-based PC. Like Citrix, Liftoff can run on older PCs such as 386s. Unlike Citrix and Prologue, Liftoff runs on a standard version of NT, but runs only 32-bit Windows applications, so older 16-bit Windows applications won't work. More of an application runs on the PC-based client, which means less work for the server. New Moon claims a server running Liftoff can handle twice the load of a comparable server running a Citrix-based solution. New Moon positions Liftoff as a complementary solution to NetPCs, claiming it reduces administration costs through centralized software version control, simplified software maintenance, and improved license management with Liftoff's integrated software license tracking feature.
A fat client is a traditional PC, high-end workstation, or laptop. Fat clients can operate attached to the network or independent of it. The advantages are performance, flexibility, and mobility. Unfortunately, fat clients are difficult and expensive to maintain.
The biggest change in the next year will be the addition of the managed PC features. When these features are better defined and prove themselves in the market, they likely will become a requirement for corporate IS. Some interesting partnerships will result as hardware manufacturers scramble to build or buy client-side management software to turn their device into a managed PC.
PICKING THE BEST
Regardless of your client strategy, your server and network strategy will need to include fault-resilient features. Lean and thin clients do not work when the server or network is down. You must factor in the cost of clustering, UPS, RAID, and key redundant components as future options.
What client strategy is best? It depends on what functionality you need. The TCO issue will increase the market for thin and lean clients, but the market for laptops and other mobile devices will also grow as companies implement virtual offices. You may end up with some of each client type in your organization. What's the best fit for each client type? Part 2 of this article will answer that and other questions, in our continuing coverage of enterprise client strategies.
Citrix Systems * 800-437-7503|
Dell Computer * 800-999-3355
Exodus Technology * 800-756-7065
IBM * 800-426-333
Insignia Solutions * 408-327-6000
Intelligent Graphics Corporation * 800-866-5597
Maxspeed * 415-856-8818
Microsoft * 800-426-9400
Network Computing Devices * 415-694-0650
New Moon * 408-557-8500
Oracle * 415-631-4600
Prologue Software * 33 1 69 29 39 39
Sun Microsystems * 800-786-0404
Tektronix * 800-547-8949
Wyse Technology * 408-473-1200