The first wave of thin-client computing kindled by Citrix in 1996 was driven by a desire to cut the hardware cost of desktops. At the time, a typical thin-client machine cost about £500 compared with £1,000 for a fat PC, but now that differential has been eroded substantially, and the emphasis has switched to total cost of ownership (TCO).
The component of the TCO with greatest cost-cutting potential is no longer desktop hardware, given that the price of standard PCs has fallen to about £500, but the cost of supporting it. This cost can be broken down into helping end-users with their problems and performing hardware and software upgrades. The latter can be substantially reduced by adopting a thin-client model, because the hardware and software upgrades are now only required on a relatively small number of servers rather than every client PC. Thin clients are, in effect, sealed boxes that will never have hardware upgrades. It is true they still run software, such as browsers, but this can all be upgraded in a controlled operation without having to visit each desktop.
As for the helpdesk side this is still needed, but the source of software problems will now be at the server end, which is a controlled environment within easy reach, and again there are fewer elements to deal with.
There is a caveat to all this, which is that the threat by thin clients galvanised the fat-client movement, and Microsoft, into action to reduce the TCO and cost of support. The proposition was that you could still have fat clients with their ability to run personal productivity applications independently of the network, but with centralised administration and delivery of applications. These applications would be maintained and upgraded on servers, but downloaded to PCs when end-users needed to run them. Furthermore, some shared enterprise applications developed for the Windows environment, such as customer relationship management, could be run centrally with PCs in this case operating in thin-client mode just handling user interface functions.
This hybrid model does reduce support costs, and even suppliers of thin-client hardware admit that the potential for further savings is then reduced. "For a company where there is a well managed PC environment, savings will be less," said Eva Lilljegren, managing director for thin-client hardware supplier Boundless Technologies.
But genuine thin clients still have the potential for further cost savings, because they lack moving parts, and cannot have software loaded into them, which itself can be a source of support problems.
Jane Rimmer, Citrix marketing director, says a significant proportion of the savings follow from preventing users doing damage by loading software and data into their PCs. This damage can only be prevented by use of thin clients.
As a substantial proportion of the support savings - probably more than 50% - can be achieved just by adopting a centralised software model with many core applications running on servers via Citrix or some other multi-user version of Windows, some enterprises use this as a migration towards a full thin-client model. In any case, not all applications, and not all end-users, are suitable candidates for thin clients, as some require the ability to run locally. This includes power users running applications that would use too much processing power and memory on a central server shared by many end-users with less demanding requirements.
These were all considerations for the publisher Harper Collins, which is currently migrating towards a thin-client model for some of its users, but meanwhile is using its existing ageing population of PCs.
Hilda Brakespeare, IT manager at Harper Collins, says an important step when implementing thin clients is to identify clearly the suitable candidates. "We divided our users into those with core vanilla 32-bit business applications and put them on to thin client, and those doing more intensive work and put them onto fat clients running NT on the desktop."
Huge savings have been made, says Brakespeare, although she is coy about precise figures. "We're aware that we recouped the investment in two years," she says.
In this case most of the end-users still have conventional PCs of some sort, but run in thin-client mode with all the applications running on central servers. "In effect, they're thin clients," observes Brakespeare. "Windows 95 is running in the background, but they can't see it and get a thin-client session straightaway."
The move to thin clients in this case was prompted by a separate decision to migrate from the increasingly unpopular Novell Netware to NT. "We realised that it would be quite expensive to replace just about every PC," says Brakespeare. "We had a mixed client base, including Macs and Suns, wanting to access NT applications, and the solution was to implement Citrix Metaframe to centralise deployment and management. In future, users will be given Windows terminals because we've run out of PCs now," she said.
Another site that opted for a thin-client solution based on NCD hardware is the London Fire Brigade. The most visible saving was on hardware, to the tune of £100,000 a year, but the primary motive was to cut the cost of supporting desktops at 120 sites around London, with 1,800 users. Previously, support staff had to visit the sites to fix problems and perform hardware and software upgrades. An interesting aspect of the project was the consideration of the server Windows multi-user software in the context of support costs.
Citrix was considered, and found to offer strong load balancing of processes between multiple servers, and also efficient communication between servers and clients via the Citrix Independent Computing Architecture protocol. But these features came at too high a price both financially and, crucially, in terms of the anticipated complexity of support. It was noted that Metaframe replaced many standard components within the underlying Microsoft Windows Terminal Server (WTS) software, motivated by the laudable desire of improving efficiency, but at the cost of making it difficult to ascertain where Citrix's responsibility ended and Microsoft's began. The fire brigade anticipated bucks being passed between the Microsoft and Citrix helpdesks, and that this would push up support costs.
NCD's Thin Path server software was chosen instead, partly because it had a clearer demarcation with WTS and partly because its load balancing was deemed stronger.
This led Bloor Research to conclude in a report on the London Fire Brigade installation, that "Thin Path could represent a more cost-effective option than Citrix Metaframe in a pure Windows environment where there is a desire to preserve the integrity of the Microsoft platform".
The report made clear that Thin Path was not necessarily superior per se, but that it did things differently and would be better in some circumstances, and should, therefore, be considered. This is reassuring because it shows that there is at least one viable alternative for the server piece of thin-client computing, which should help to keep costs down for a technology whose primary motive is to save money.
Bloor concluded in its report that the adoption of thin-client technology has led to a 15% to 20% annual reduction in direct IT costs at London Fire Brigade. This is modest compared with the claims of some thin-client suppliers, but Bloor also cited other support-related benefits. Centralisation of operations has resulted in improved quality and consistency of service and fewer "version-related" file sharing problems for users. Bloor also found that end-users had become more productive because they spent less time troubleshooting their own problems.
So far, most of the thin-client movement has focused on the Windows environment, but recently it has spread to the Unix world as well. Sun has been particularly active with its Sunray terminal, or Internet appliance range.
According to Sun's thin-client product marketing manager, David Clewes, similar TCO savings can be made in the Unix environment, but with the accent perhaps more on the capital costs. For example, you can buy a single entry -evel Sun Unix server capable of supporting 20 Sunray thin clients for £2,500. The software for a single CPU costs £300, and the entry level Sunray 1 costs £200. The cost, therefore, of a thin-client Unix set-up for 20 users would start at about £7,000, while the same number of Unix workstations might cost three times that.
No wonder then that sales of Sunray terminals have taken off, with as many sold in the UK during the three months to September as in the whole year before that.
So, perhaps this is the year of delivery for thin-client technology.
The thin-client story encapsulated
For the Microsoft world, the thin-client model with discless desktops and the core operating system plus the applications running on servers, was invented by Citrix with its Metaframe technology in 1996. Metaframe is a layer of software extending NT to support multiple clients from a server rather than just a single user operation.
Noting the hype generated, Microsoft's initial reaction was to thank Citrix for creating the market and then taking it over. But it became clear that what Citrix had achieved in converting NT from an operating system that was just multitasking into one that was multi-user as well was not so easy to imitate quickly, and so the two companies cobbled up a deal.
Microsoft acquired the rights to essential ingredients of Metaframe and these were incorporated in its own Windows Terminal Server (WTS), which remains a vanilla multi-user extension to NT and now Windows 2000.
Meanwhile, Microsoft cut the rug from under Citrix to some extent by changing the licensing rules for NT to take account of the fact that a single version of the operating system could now support multiple users. Until then, NT had been licensed according to the number of installations, which equated to the number of users. But with the thin-client model it became possible to license a single copy of NT and deploy it a number of times to users whose thin-client machines did not run the operating system. Microsoft closed the loophole by matching the licence to the number of users, rather than machines.
This, as David Perry, worldwide director of marketing for thin-client supplier NCD, noted, reduced the basic capital cost saving that could be achieved by going for thin clients, because more had to be paid for the Citrix software. "This combined with the reduction in entry level PC costs [fat clients] has reduced the disparity between PC and thin-client costs significantly," said Perry. Some disparity remains, but the emphasis is now firmly on the total cost of ownership (TCO) savings.
While Citrix continued to dominate Windows server-based computing as realism replaced the hype between 1997-99, there were a few other contenders for the server piece of the action, notably NCD with its Thin Path. Both NCD and Citrix use WTS for the basic multi-user functions, but enhance these with additional features, notably load balancing, so that the processing can be distributed across multiple servers to improve performance and availability. Citrix stuck to its knitting-in software, but NCD also makes thin-client hardware, for which it is now probably better known. Others such as Wyse and Boundless majored on the thin-client hardware.
A more recent development has been the ability to access the server-based Windows applications from any browser via, for example, Citrix' Nfuse technology, as well as conventional thin clients.
Another recent development is extension of the action to the Unix world. Unix has always had multi-user operation, but now Sun has its Sunray range of thin clients, which it calls information appliances, supporting a more comprehensive range of GUI options, such as smartcard access. These offer a thin-client option for power users, offering similar TCO benefits.
The latest chapter in the story features the arrival of SCO's Tarantella initiative, which competes with Citrix, NCD and Microsoft on the server side of thin-client computing, but from a different angle. Instead of being just an extension to the Windows server operating system, it comprises middleware in between the client and the server. At the back end it communicates with the server operating system in the appropriate protocol, such as X11 for Unix, or ICA for NT, or 3270 for IBM System/390, and converts the instructions into HTML to support front access from a browser.
In this way clients could be PCs, or any other form of client capable of running a browser. This could offer potential TCO savings for larger enterprises wanting to provide access to a variety of back-end systems, not just Windows, via an intranet. But for smaller sites accessing just Windows, Citrix, NCD or indeed Microsoft seems more cost effective.
Potential cost savings in a nutshell
(These savings will be irrelevant if the thin-client model is deployed for the wrong applications, or for end-users whose productivity is compromised. Good implementation is more critical, because with server-based computing problems are amplified to a larger number of users.)