Linux Software Stacks

Linux is regularly appearing on IT organizations' shortlists for Web and application deployments. It is increasingly being considered for small databases (2-4 processors). In a Lintel world, we are now seeing the extension of open source software not just with the Linux OS but also for open source solutions.

META Trend: Through 2007, widening annual platform price/performance improvements (z/OS at 10%-15%, Unix/RISC at 30%-35%, and Wintel/Lintel at 35%-40%) will increasingly drive server market segmentation. Linux's rapid maturation will initially be at the expense of Unix (2004/05), but will eventually challenge Windows (2005/06). By 2005/06, Unix will continue to recede to back-end, legacy status, driven by distributed n-tier server architectures and compelling Intel performance and volume economics. By 2007, z/OS's high software costs and anemic annual hardware price/performance improvement will slow zSeries' annual net capacity growth to 10%. Increasingly stable, cost-optimized legacy platforms (e.g., z/OS, OS/400) will become attractive outsourcing candidates through 2005/06.

Through 2008, most software services and applications should be provisioned with a choice of either commercial software or open source alternatives. Measurement of the capabilities of open source alternatives is often a soft area of focus, with each IT organization (ITO) having different values and criteria for the attractiveness of Linux and open source solutions. A majority of these solutions will be on Linux on Intel (Lintel) based platforms with corresponding service software. Through 2006, commercial software offerings on Linux will be more attractive across the stack than a pure open source alternative, due to the high risk and cost of integrating and supporting complex “freedom of choice” open source solutions.

We believe ITOs should maintain at least one piece of commercial software within a Linux or open source stack and allow at least one company to be responsible for complex service support and integration issues. Otherwise, they must understand the risk of such stacks and scrutinize open source software and applications with the following criteria:

• Service: The service is the description of the open source software function, and includes Web server, application server, database (DBMS), file and print services, directory services, management tools, and operating system or Linux OS capabilities.
• Application: This is the group or class of Linux applications categorized by research/high-performance computing applications (HPC), packaged applications, homegrown J2EE (Java) applications, and proprietary applications.
• Risk: Risk is defined here as a reflection of maturity and complexity that impacts the service level across the ITO’s portfolio of individual applications and services.
• Maturity: Maturity is a measure of proven stability, stable/rising market share, broad vendor commitment, and the extent of (vendor) support available.
• Complexity: The measure of complexity includes a high degree of integration, multiple interdependencies (e.g., systems administration/management), the internal know-how required, and requirements for skill transfer, depending on the responsible location and high costs and efforts for rollout/installation.
• Recommendations/comments: These are the top-line recommendations (or related comments) we have for Linux applications and services.

After gaining understanding of these measurements, an ITO can then assess the capabilities of the open source service. Our research indicates a diverse range in these services across the Intel criteria (see Figure 1). We position open source services as follows:

• Web server: Web servers are a commodity, almost appliance-like in their single function. There is a simple choice for the open source world: Apache or Microsoft. Any other Web server platform is a niche platform. Application server: Application servers demand more resources from the operating system and the platform, but, again, these are near commodity, and Linux platforms running JBoss for non-complex workloads should be considered.
• DBMS server: Linux scaling as a DBMS platform will be restricted to an Oracle or DB2 scale-out model until the introduction of the 2.6 Linux Kernel, which will have increased scaling, threading, and I/O capabilities benefiting DBMS deployments. Open source DBMS software (e.g., mySQL, maxDB) is now being considered for small DBMS workloads.
• File and print: File provisioning is best fulfilled by network-attached storage or a storage-area network within an enterprise service portfolio. These services can be fulfilled by Linux and open source solutions, but they are immature and dependent on directory services in either an open environment (DNS and LDAP) or Microsoft Active Directory.
• Directory: Directory services are not well established on Linux or for open source in complex enterprise deployments. These services will mature through 2005, following Linux kernel changes that will enhance parallelism and I/O scaling.
• Management tools: This is currently the most immature area of Linux. To coin a phrase, “Linux and open source management across the stack is as good as a C:> in DOS!“ ITOs considering Linux and open source must invest in separate services for infrastructure management, such as adaptive resource management (ARM) tools.
• Linux OS: Linux itself is mature enough for most deployments (other than complex DBMS workloads of more than four processors that require the 2.6 kernel enhancements).Through 2005, we expect SUSE with Novell and Red Hat to extend the functionality of its shrink-wrapped Linux offerings to include more commercial service software as well as open source solutions.

Having understood the software services capabilities of Linux and open source, an ITO can use the same criteria to measure how appropriate Linux is for different types of applications (see Figure 2). We position Linux application capabilities as follows:

• Research and HPC applications: High-performance computing and research/technical applications are well suited to Linux, where the rich development environments and tools help benefit the versatile performance attractions of the Lintel platform. HPC is best deployed on Linux clusters.
• Packaged applications: Packaged applications (e.g., SAP, Oracle, PeopleSoft, JDE) are moving Linux deployments to development platforms or primary ports. The risk for an ITO here relates to the application vendor’s support of Linux and how aggressive the vendor is in sales, marketing, and support. As packaged applications on Linux increase, the increased presence will help vendors attain a larger share of the total project portfolio.
• J2EE (Java) applications: Java-based homegrown applications are not as easily considered on an open source stack. Complex J2EE applications using BEA or IBM extensions, when moved to Linux platforms, are best kept on BEA or IBM respectively. Simple J2EE applications can be considered on JBoss, but support or integration costs may offset any initial license savings. When considering J2EE applications for Linux deployment, a reflection of overall costs is fundamental.
• Proprietary applications: Applications written for a specific platform (e.g., Unix), with complex scripts or direct hardware access, are the least likely candidates for migration to Linux. Here, the skills and resources of the source (Unix) application ISV and the target Linux deployment are important for migration. Groupware/messaging applications: Packaged groupware and messaging applications such as Lotus Notes follow the same guidance for packaged applications, subject to support and commitment to Linux from the individual vendor.

However, some open source products overlap with these productivity suites (e.g., Open Office) and are a distraction. Through 2006, these open source tools will rarely make it into mass commercial deployments. As they mature, the open source tools will become attractive for small and medium businesses and gain increased enterprise acceptance.

Bottom Line: When selecting Linux services or applications, a uniform set of criteria should be considered in a non-uniform open source environment. IT organizations must balance the relative cost of development and integration of open source solutions with that of canned, proven solutions on commercial stacks.

Business Impact: Selective provisioning of Linux applications and open source services helps to optimize an IT organization’s portfolio.

META Group originally published this article on 24 February 2004.