Users face considerable vendor hype, especially from IBM, concerning the viability of Linux as a mainframe growth driver. Although users have been slow to embrace mainframe Linux - despite IBM's marketing claim that 11% of MIPS shipped in 2001 were Linux-ready - we see some near-term Linux momentum developing for application integration with existing legacy systems, operations training, and low-end consolidation (largely driven by limited staff resources).

One of the primary drivers of Linux's generic growth across all platforms is the fundamental distrust by users of the Microsoft Windows 2000 juggernaut that we expect to dominate enterprise data center computing by 2007.

Our research indicates the modest user adoption of mainframe-based Linux is being driven by the mainframe's current ease-of-use/management lead versus current incarnations of Unix and Wintel-based systems. The mainframe excels in relation to partitioning, workload management, reconfiguration, restart, prioritization, and general VM ease-of-use characteristics. Indeed, the main attraction of mainframe Linux is as a simple "sandbox" for user experimentation, and to more closely connect existing applications compared to outboard Win2000 or Unix solutions. Mainframe-based Linux's near-term attraction is its ease of operation, not its price/total cost of ownership.

We expect mainframe Linux to remain a niche, whose term will be defined by mainframe/Wintel gaps of cost (widening) and management functionality (closing). Largely limited to leftover (i.e., not new) MIPS, mainframe Linux's near-term "sweet spot" will be applications that complement/enhance legacy delivery. Our research indicates few users are buying significant new mainframe capacity to run Linux.

Longer term (2005-07), as Unix/Win2000-based systems flesh out increasingly robust, mainframe-like management capabilities (e.g., partitioning, workload management, reconfiguration, prioritization), the justification for paying the mainframe's significant cost premium will fade. Indeed, by 2007 the 15%-20% of Linux-based data center workload will run almost exclusively on Intel, which by then will have reasonable partitioning and workload management.

Moreover, we do not believe IBM will be able to match the aggressive 35%/year price/performance improvement of the mainframe's platform competitors. This will put further economic distance between the mainframe versus Intel as users' Linux platform of choice. Even IBM's new, relatively low-priced Linux-based z800-0LF hardware remains 8x-10x more costly than Intel-based systems, and 4x more costly than Unix. This is in the context of the mainframe's sluggish (less than 20%) current and projected annual growth relative to its Unix and Win2000 competitors (50%-60%).

User Action: Users should view the mainframe as an interim and tactical port in the early Linux storm. Near-term Linux mainframe deployment (through 2004) should be limited to consolidation of simple, non-critical applications, and to gaining Linux expertise in a controlled, low-risk environment. Longer term (2005+), we believe the Linux center of gravity will inexorably migrate to the compelling economics and mainframe-class management capabilities of Intel-based blade servers, leaving little justification for a continued mainframe Linux presence.

We caution users that current Linux incarnations are relatively immature, as evidenced by the interminable list of errors/patches on Linux providers' Web sites. They have, therefore, been naturally limited to simple, less-critical applications such as LL Bean's e-mail notification system, which is mainframe co-resident with the application. Indeed, our research indicates that controlling/managing "version churn" is an increasing Linux burden, with many users turning to vendors such as SuSE to keep Linux and future applications synchronized.

Perhaps the most important aspect of mainframe Linux is as a tool to keep other mainframe costs down. Users should leverage the significant (3x+) disparity between IBM's aggressive pricing incentives for running Linux on the mainframe against legacy workloads running on the identical hardware platform. Indeed, users acquiring low-cost ($1.6K/MIPS) Linux-dedicated mainframe capacity (e.g., the z800-0LF) should, at minimum, insist on the contractual right to use those MIPS for legacy workloads in a disaster recovery scenario. They should also request the flexibility to shift such Linux MIPS to legacy workloads at no additional charge, as business requirements dictate.

META Group analysts Rich Evans, Rob Schafer, William Snyder, Brian Richardson, Philip Dawson, Rakesh Kumar, Dale Kutnick, David Cearley, and Jack Gold contributed to this article.

Send a comment or question about this article