The Real Value of Linux
Under the guise of reducing total cost of ownership (TCO) in a tough economic environment, technical staffs are recommending replacement of Windows with Linux on their servers. This is based on the flawed assumption that because Linux is "free" it will reduce TCO. On closer inspection, it appears the recommendation is more an emotionally driven reaction against Microsoft than a factual case for Linux. Astute IT organizations will recognize that Linux's true value is derived more from the price/performance of the commodity Intel hardware it enables than from its open source characteristics.
META Trend: With highly distributed n-tier (DBMS, application, Web) server architectures commoditizing during 2002-04, Unix (other than Solaris) will recede to high-end, low-volume, niche-platform status by 2005/06. Windows will increasingly dominate for midtier application servers (2002-04), due to growing ISV reference platform momentum, and be a suitable DBMS server for 90%+ of application requirements. Linux on Intel ("Lintel") will be a successful high-volume Web, technical computing, and appliance server OS, but enterprise application package (2003/04) and DBMS (2005/06) server penetration will be slower. Linux software and services prices will increase to about 10%-20% less than those of Windows by 2004/05.
Our research indicates a strong interest in using Linux in the data center, but few clients understand the real value of Linux, nor have many clients embarked on major Linux projects outside of Web server farms, appliances (network-attached storage [NAS]), or general infrastructure servers (e.g., DNS, DHCP). Through 1Q03, Linux will begin to penetrate the application server tier, with IBM and BEA targeting Linux on Intel as a low-cost J2EE platform. By 1H03, Oracle Real Application Clusters (RAC) will demonstrate adequate high-availability clustering capabilities, enabling Linux to begin penetrating the low-end enterprise database market (i.e., 2-4 CPUs - see Figure 1). By YE04, Lintel will constitute 25% of server sales for the application server tier and 10% of server sales in the database tier. By YE07, this will increase to 40% and 25% respectively.
Although Linux has established a foothold in the Web tier due to the popularity of the Apache Web server, it still ranks a distant third (10% of Fortune 1000 companies) behind Solaris (30%) and Windows (50%). By 2007, Linux and Windows on Intel (“Lintel” and “Wintel”) will be the dominant platform for the application server tier, leaving RISC/Unix vendors competing with IBM mainframes in the high-end database server tier (i.e., more than 8 CPUs). The success of Linux will come primarily at the expense of Unix (e.g., Solaris, AIX, HP-UX) rather than Windows servers. Indeed, the true value of Linux is as a viable alternative server operating system (OS) for commodity Intel hardware.
Linux has emerged as the darling of the "technical crowd," yet the interest appears be more emotional than factual, and is based on a questionable lower-cost-of-ownership argument. The Linux OS license is “free,” but that does not ensure that total cost of ownership will be reduced. For example, Linux requires more staffing resources and effort to match the reliability, availability, and scalability of high-end Unix and Windows 2000 or XP servers. Users must purchase high-availability add-ons (e.g., clustering partitioning, “journaled” file systems) and support from third parties, which increases cost and complexity. Through 1Q04, this will limit Linux use to applications that do not demand high levels of reliability, availability, and scalability in a single server, such as Web server farms, application server farms, non-mission-critical applications, and, as Oracle RAC matures, DBMS clusters.
Even if all other Linux costs were the same, the impact of its free OS license on total cost of ownership of a significant project (e.g., ERP, CRM) would be minimal, because the OS license fee typically is less than 2%-3% of the TCO. It is only when other significant pieces of software can be licensed at little or no cost (e.g., office suite, e-mail, DBMS) that TCO reduction is at a level significant enough to merit the additional complexity, risks, and potential cost overruns of Linux.
The key attractions to Linux are:
- Royalty-free distribution: This is clearly an attraction, making Linux essentially free, yet packaged versions that include documentation and support are not free of charge . We project that, through 2005, Linux prices will rise to within 20% of the cost of Windows.
- Access to source code: All versions include source code, making Linux compelling for technical staff. Governments and security agencies like the ability to check for backdoors and security vulnerabilities in the code.
- High levels of reliability: Based on the Unix architecture, and after extensive testing by the open source community, Linux has proven to be surprisingly robust as a new OS. Although this was compelling compared to NT 4, increased stability of Windows 2000 has narrowed this gap, making this less of an advantage. Yet Linux is still missing native high-availability features such as journaling file systems or clustering.
A: Linux has its place in the data center, but it is not a silver bullet for Windows. Linux can be used in the following instances:
- In an appliance where the OS is not exposed (e.g., NAS, kiosks). Linux removes the licensing cost issues, and the appliance maker becomes responsible for the support and integration of Linux. Because the OS is not exposed, it does not have an impact on staff skills and training.
- When the price/performance benefits of commodity Intel servers over RISC are important, but single-image high availability is not an issue. Intel servers are widely used for scientific computing due to the computational price/performance of the Intel chipset. Web and application servers usually are computed bound, and Intel-based servers can deliver 2x-3x better price/performance than RISC. Although it is possible to use Windows in this application, many Unix-centric organizations will be more sympathetic to Linux and will find the skill transition much simpler.
- As a general-purpose infrastructure server (e.g., DHCP, DNS, POP), where solid reliability is required but high availability is not. Linux is particularly well suited in this role, because it is robust and many infrastructure products are readily available.
Q: “Can I use Linux to replace Windows for file and print?”
A: Although this is possible using Samba (an open source file-and-print service for Unix), it is not recommended. To simplify management, file and print must be incorporated in the directory services strategy that is used for the entire desktop infrastructure (e.g., Microsoft’s Active Directory or Novell’s ZENworks). Samba has basic access controls for the file and print services, but it does not replace a directory service, nor does it integrate with Active Directory of ZENworks in its current release (Samba 3.0, currently in development, will incorporate this).
Instead of using Linux to replace Windows file and print servers, IT organizations should consider either consolidating the number of servers or using a NAS server for the file sharing. We have yet to see significant projects that replace Windows with Linux for file and print. A switch to Linux for file and print might lower purchase costs, but it would seriously affect the ease with which users can access the services as well as increase management complexity, thereby driving up the total cost of ownership.
Business Impact: Inappropriate use of Linux as a Windows or Unix replacement will weaken the IT infrastructure and reduce its business value.
Bottom Line: Organizations that allow emotional reactions (e.g., against Microsoft) to drive decisions to replace Windows or Unix with Linux will fail to achieve anticipated savings, and will end up with an infrastructure that is limiting and difficult to manage.
META Group originally published this article on 2 December 2002.