RFG believes enterprises must consolidate data center servers to enterprise-class servers such as a mainframe or a Superdome to satisfy the business demands for greater availability, enhanced business continuity, improved flexibility, energy efficiency, cost reductions, and increased productivity.
Consolidation and virtualization are mainstream concepts today as is the desire for energy efficiency. As part of these initiatives the notion of moving Linux from x86-based servers to a mainframe Integrated Facility for Linux (IFL) or an enterprise server is an option being pursued by many enterprises.
But before any target platform is selected, IT executives should carefully evaluate whether an IBM mainframe IFL or an enterprise server such as the HP Superdome is the best business choice for the applications being re-hosted.
Top 10 Linux on Mainframe or Alternative Considerations January 2008 Cal Braunstein Chairman/CEO and Executive Director of Research Robert Frances Group 120 Post Road West, Suite 201 Westport, CT 06880 http://www.rfgonline.com Untitled Document Robert Frances Group, Inc. 2 RFG believes enterprises must consolidate data center servers to enterprise-class servers such as a mainframe or a Superdome to satisfy the business demands for greater availability, enhanced business continuity, improved flexibility, energy efficiency, cost reductions, and increased productivity. Consolidation and virtualization are mainstream concepts today as is the desire for energy efficiency. As part of these initiatives the notion of moving Linux from x86-based servers to a mainframe Integrated Facility for Linux (IFL) or an enterprise server is an option being pursued by many enterprises. But before any target platform is selected, IT executives should carefully evaluate whether an IBM mainframe IFL or an enterprise server such as the HP Superdome is the best business choice for the applications being re-hosted. To do a thorough evaluation one needs to understand the application solution requirements and then map them to the characteristics of the target environment. This must be done on a global basis and not just for a single application, as an enterprise server is expected to handle effectively all applications it is executing. In this report RFG will only compare the HP Superdome against the IBM System z with an IFL as the Linux host, since both are used as centralized consolidation servers. Readers interested in other platform choices should be able to extrapolate the factors to other systems. There are 10 different factors that need to be examined to see which platform base is the best choice for consolidation of Linux x86-based applications. These determinants are as follows: 1. The platform: Linux only or Linux and other operating systems such as Microsoft Corp.'s Windows and Unix. 2. Application performance characteristics a. Single large high performance application or multiple heterogeneous applications. b. Real time transactional, batch or decision support applications. c. Processor or I/O intensive 3. TCO, ROI or other financial return metrics 4. Utilization and utilization variability characteristics 5. Skill on board and required 6. Security level 7. ISV support of platform and system type 8. Energy efficiency 9. Manageability and scalability 10. RAS: reliability, availability, and serviceability Some of these factors are entry requirements for any enterprise server solution while others are advantages of one platform over the other. 1. The platform. Enterprises interested in consolidating only Linux applications can take advantage of either platform choice. The System z IFL is strictly a Linux processor and is designed to handle 54 Linux images natively or up to thousands of virtual machines when the z/VM operating system is employed. However, if the objective is to consolidate Linux, Microsoft Corp.'s Windows, OpenVMS, and Unix applications onto the same enterprise server, then the HP Superdome is the only choice. The Superdome supports up to 128 Linux virtual images natively or more than 1000 virtual images with Integrity VM on a single server while allowing a mix of applications with different operating systems running on the same server. Executives need to examine this issue holistically, as most organizations have a preponderance of Intel servers, usually Windows-based, that seem like they are growing daily. Executives need to understand the mix of operating systems on these servers and determine what the target environments e.g., blade servers, commodity Intel servers, mainframes, or Superdomes for these operating systems and applications should be. 2. Performance. The application performance characteristics should be analyzed from three different angles. The first is to see if there is a single large high performance application running on a number of Intel servers that can be migrated and needs to be on its own consolidation server for availability, mission-criticality, or performance reasons. Or if there are multiple heterogeneous applications that are to be run on the enterprise server and while they each may have unique performance characteristics, they are independent applications from a performance standpoint. Each of the applications can be classified as batch, decision Untitled Document Robert Frances Group, Inc. 3 support, or real time transactional. They may also be processor or I/O intensive. The mainframe IFL sweet spot is multiple transactional, non-CPU intensive Linux applications. For a large single application like an ERP application that may be the only application on the server the Superdome proves to be the better performer. The HP Superdome can also support multiple Linux applications, albeit somewhat less efficiently than the mainframe. The Superdome's scale-up architecture makes it very strong in running a smaller number of performance-sensitive, larger, more complex Linux images. For these mixes or in fact, any other application combination being ported to an enterprise server, executives should consider piloting the workload, as there are too many variables to make a definitive statement that for all possible conditions the mainframe or Superdome as the best performing solution. 3. TCO, ROI or other financial return metrics. RFG finds that a total cost of ownership (TCO) analysis works best here. One of the big variables in this analysis is the requirement for an enterprise to purchase a general purpose System z mainframe so that they can acquire and use the IFL. A mainframe, no matter how small, is not inexpensive and to have it totally underutilized is not cost effective and a waste of resources. Moreover, while an IFL costs much less than a standard mainframe, it still lists in the 95,000 range. Additionally, executives need to be sure they factor in the utilization levels into the evaluation. That is, most single application x86 servers have a utilization of under 15 percent (usually less than 10 percent and half of that is not being used by the application) and x86 blades run at under 60 percent utilization. On the other hand, enterprise servers and mainframes should be evaluated as loaded to operate at an 80 percent utilization level as their flat running rate. Also to be factored into the equation should be the reduction in administrators or operators required once the applications are consolidated. Administrators/operators may require cross-training to reach the utilization levels attainable when systems are consolidated on an enterprise server. The analysis should also include the environmental factors as energy efficiency now plays a significant role in the cost savings. Software costs are another major factor that needs to be carefully evaluated. The Novell, Inc. basic SuSE Linux Enterprise Server price for the HP Integrity line is 349/server while the price on an IBM System z is 11,999. Since consolidation may result in a number of copies of other products being in use on the same server, executives, working work with the involved vendors, should determine the pricing impact on the platform. 4. Utilization and utilization variability characteristics. The objective in an enterprise server is to operate at an 80 percent steady state utilization level, with bursts that could drive the utilization up to 100 percent or more, if additional processing capacity existed. HP and IBM have addressed this issue with a feature called capacity on demand (CoD). While the name is the same, the implementation of the CoD feature varies significantly and can have a huge impact on overall operating costs. HP's CoD is more flexible in that it offers global capacity usage rights across servers and a significantly finer billing structure that helps to contain temporary capacity usage costs. This difference, which should be factored into the TCO analysis, could be quite costly if the enterprise hits frequent temporary periods of excess usage. In addition, HP includes in its CoD feature site installed and pre-configured cabinets, connectors, memory, power supplies, processors, and storage while IBM's CoD offers site pre-installed memory and processors. 5. Skill on board and required. Companies without mainframes installed need to determine the impact on resources and skills caused by the inclusion of a new processor type with its unique operating systems and operating environment. New operating systems require new skills even if the intent is to only run Linux on the IFL. Existing staff will need to be cross-trained or new staff hired. The efficiency and effectiveness of this environment will depend on the overall usage of the mainframe and the staff skill mix. On the other hand, consolidation onto a Superdome will be easier to adjust to and implement since Linux commands and usage is similar to those found in the HP-UX operating system. Untitled Document Robert Frances Group, Inc. 4 6. Security level. For most business application settings there is no discernible difference between the two systems. In fact, Superdome access control, certification, encryption, and intrusion detection will satisfy all but the most demanding certification applications where EAL5 certification is called for. 7. ISV support of platform and system type. Executives need to decide whether or not the applications being ported to the mainframe or Superdome require the ISV to support it on that platform. The issue is the maintenance of the application by the vendor when an application failure occurs on the production system. If the application is not running on an approved or certified system, the support commitments may be invalidated and rapid correction of the problem may become problematic. 8. Energy efficiency. Energy efficiency is an element that must be addressed in any platform consideration. Unfortunately it is not a simple one to evaluate. If one measures it solely by the energy consumed (kWatts) or heat given off (kBTUs) by a mainframe or Superdome, then depending on the configuration the Superdome would outperform or equal the latest System z. However, the correct way to measure the energy usage or heat generation is to do it on an application portfolio basis. That is, each server should be loaded with the desired applications on systems that have been configured to run at 80 percent utilization for the test environment. Then the energy efficiency of those two configurations should be measured. For applications that are best suited for a Superdome (see tip #2), the energy efficiency of the server should also be better than that of a comparable mainframe configuration. 9. Manageability and scalability. Both platforms provide similar capabilities for partition, print, system, virtualization, and workload management and both are highly scalable. RFG views that these features will be a wash for most applications and not a differentiator for the majority of enterprises. 10. RAS: reliability, availability, and serviceability. RAS has traditionally been viewed as the realm of the mainframe. However a Superdome's reliability is comparable to that of a mainframe and it can achieve the same levels of availability that one expects from a mainframe if properly configured and managed. Lastly, Superdome hardware and software support, service notification and upgrade capabilities yield outage impacts from system failures and upgrades comparable to a System z. In effect, RFG finds that RAS is no longer a mainframe differentiator. RFG believes the key selection criteria for a Linux consolidation platform are application characteristics and mix, current and target operating system bases and associated support resources and skills, ISV support needs, TCO, and utilization characteristics. The security certification requirement will only play a role in a very limited set of circumstances and all other criteria should be secondary. While IBM's mainframes have the history of being the enterprise-class consolidation and virtualization server, the Superdome is a current and future contender as the Linux consolidation server. IT executives should carefully evaluate the choices before deciding which platform best satisfies the business requirements for availability, business continuity, flexibility, energy efficiency, cost savings, and productivity. HP commissioned the Robert Frances Group to conduct independent research and analysis of this business issue. The content of this report is the product of RFG and is based on independent, unbiased research. Although every effort has been taken to verify the accuracy of information herein, neither RFG nor the sponsor of this report can accept any responsibility or liability for reliance by any person on this research or any of the information, opinions, or conclusions set out in the report. Copyright 2008 Robert Frances Group, Inc. All rights reserved. 120 Post Road West, Suite 201, Westport, CT 06880. Telephone 203/429-8950. Facsimile 203/429-8930 www.rfgonline.com. This publication and all publications may not be reproduced in any form or by any electronic or mechanical means without prior written permission. The information and materials presented herein represent to the best of our knowledge true and accurate information as of date of publication. It nevertheless is being provided on an "as is" basis. Reprints are available for a nominal fee.