Server platforms Toolkit

The march of Linux in the enterprise

George Weiss, Edward Younger Gartner

Published: 11 Mar 2004 14:45 GMT

Server Operating System Growth
Strategic Planning Assumption: Only three growth server operating systems (Windows, Linux and AIX) will exist through 2008 (0.8 probability).

Gartner expects several key market share trends through 2008:

Through 2008, the server operating system shakeout will continue, and the server microprocessor shakeout will accelerate. As a result of these shakeouts, Gartner expects the following:

– Only three growth server operating systems (Windows, Linux and AIX) will exist through 2008 (0.8 probability).
– Only three growth server microprocessor architectures — IA-32, Itanium Processor Family (IPF) and Power — will exist through 2008 (0.8 probability).

Linux will increase market share significantly, much of it from the use of scale-out architectures such as blades, compute clusters, replicated systems and appliance-type deployments.

By year-end 2008, AIX will increase market share and approach or pass Solaris as the No. 1 Unix operating system (0.8 probability).

The IPF architecture will increase, but slowly. IPF server revenue will not cross the $1bn revenue mark until 2005 (0.8 probability).

The Linux Road Map
Strategic Planning Assumption: Improved scheduler, memory management and threading will add a 20 percent to 30 percent performance boost in the new version of Linux 2.6 (0.8 probability).

The Linux road map is not locked to specific milestones. It depends on collaboration from OSS projects, hardware and software vendors, the Linux distributors and Linus Torvalds' kernel organisation. A rough estimate of some of the upcoming enhancements shows increased performance improvements through kernel efficiencies, such as queue scheduling, thread handling, asynchronous I/O and clustering with failover.

These enhancements should result in scaling with demonstrated production performance to eight-way symmetric multiprocessing configurations in 2004, as well as eight- to 16-way systems in 2004/2005.

These improvements could open the way to wider Linux adoption, but scaling won't be enough. The operating system must have the proven reliability characteristics of current Unix, mainframe and other proprietary operating system environments before many data centre managers will feel comfortable.

For example, the operating system must provide error detection and diagnostics, work reliably supporting multiple and mixed applications, and have good recoverability and error handling from logical hardware problems.

The Linux operating system must have longer release cycles with stable version levels for ISVs to minimise their maintenance and upgrade costs. Therefore, it is more likely that Linux will continue to have predominant adoption in multinode and replicated deployments, with a longer adoption cycle in back-end, mission-critical applications.

Action Item: Consider Linux for its strength in horizontal clusters, but avoid replacing robust back-end systems until further progress is made on its evolution in the 2005 to 2007 time frame.