For years, server consolidation has been touted as beneficial by vendors, clients, and analysts alike. Yet our research indicates that many consolidation projects have been overly consolidated, resulting in negative ROI - or increased costs! IT organizations need be realistic about consolidation savings and see beyond vendors' self-serving proposals for potential total-cost-of-ownership (TCO) savings.
META Trend: Through 2006, non-mainframe infrastructure consolidation will be driven by value-based portfolio management principles, but will be hampered by tool immaturity, premium-priced high-end servers, inflexible software pricing, and chargeback politics. Infrastructure co-location and networked storage consolidation will become standard by 2004/05. Best-practice infrastructure cost-efficiency will be defined by effective software asset and vendor management, support cost containment, and extension of server and storage refresh cycles. However, maturing server partitioning and integrated workload management will have minimal impact on staffing and total cost of ownership. When coupled with exorbitant high-end server pricing, these issues will inhibit "big box" consolidation.
Through 2010, benefits gained from server rationalization will outweigh those of server consolidation, since diminishing the number of types of things reduces cost more than diminishing the number of things. If IT organizations (ITOs) do not reduce complexity, consolidation projects will become costly failures. Consolidation itself is not a big-box “technology” phenomenon or feature set. It is a series of steps or “processes” as part of an IT program, balanced with increased efficiency and skill optimization through use of the appropriate resource management tools. Although larger vendors such as IBM, HP, and Sun all lead IT infrastructure consolidation projects with a “trust us” message, we must question that trust. As part of these consolidation programs and promised reductions, the vendors want to sell more, do less, or add more hardware, software, and services - or have a complex “trust us” mix of all three of these approaches.
META Group has been working on server consolidation models for more than five years. Our methodology has been proven and refined during that time through hundreds of client engagements, workshops, and interactions. A significant aspect of refining the model was the addition of the rationalization function to the existing co-location, storage consolidation, and workload consolidation model (see Figure 1).
Our recommendation for server rationalization is to reduce complexity by understanding, retiring, or outsourcing complex niche and legacy workloads; or alternatively, to move them to a mainstream platform. Vendors, on the other hand, see rationalization as a way to “move other vendors’ stuff” to “our stuff” - which is not rationalization, but vendor substitution. Server rationalization offers the biggest cost savings of all consolidation practices by reducing the complexity of niche and/or legacy platforms.
To gain understanding of the true business case for consolidation and potential savings, we have pinpointed five server vendor myths that are commonly used in consolidation case studies and methodologies. ITOs should consider these myths whenever a vendor proposes to consolidate another vendor’s “stuff” or offers to consolidate an ITO’s entire portfolio.
Myth #1: Consolidation = TCO Savings
Server consolidation justification is driven by cost savings of either the platform or operations. Yet when considering consolidation case studies, “apples-to-apples” comparisons must be made. The TCO benefits of many consolidation case studies are attained not from consolidation only, but by comparing old releases with new products, unmanaged servers with managed servers, and the distributed environment with the consolidated environment.
To compare a distributed, three-year-old, unmanaged, Intel 4-way Pentium Pro Microsoft NT3.5 server farm with a new Unix/RISC platform that is centrally managed is an unfair TCO example. Vendors often tout footprint reduction and server administration efficiencies of 2x-3x. If the client compared the same NT servers with a new Intel Xeon uniprocessor Windows 2003 server deployed with the relevant adaptive resource management tool, similar improvements would be gained - without the change in platform and architecture. In addition, Moore’s Law (i.e., the doubling of transistor density over 12-18 months) has resulted in the legacy 4-way being replaced by the single processor (managed) server.
Myth #2: One (Big-Box Server) Size Fits All
Resource management and provisioning used to be a feature of a systems administration console, the single point of management associated with the big-box platform. Now resource management should be treated as a service for all platforms, new and old. Blade servers can be considered for Web use and some applications, but they require the use of ARM tools. Again, these tools improve the systems administration ratios, not consolidation itself. Sixteen managed rack-based servers offer the same consolidation savings as 16 partitions on a big-box solution with significantly less hardware cost. Fewer than 10% of database workloads require big-box flexibility and scalability. Only workloads requiring highly flexible, scalable workloads (high-end DBMSs) should be considered for big-box solutions. Big-box solutions should not be considered as a consolidation platform for all workloads or for general-purpose workloads or application consolidation.
Myth #3: One Vendor/Operating System Can Do It All
META Group believes an ITO can over-consolidate and rationalize a portfolio to a single vendor. This may be acceptable if the ITO is considering outsourcing its operations, but for most situations, that is not the case. For all infrastructure services (e.g., servers, storage, database), we believe a two-vendor strategy is optimum. Primary and secondary vendors are acceptable, but we do not believe that one server vendor does offer leadership in all services. Even if this is considered a possibility, by moving to a single vendor, the short-term discounts will be offset by an over-dependency on a single vendor, and negotiation positions will be weakened over time. We also strongly recommend consolidation of operating systems. Moving to two operating systems (Windows and/or Linux) for all new Web and application servers and rationalizing the DBMS platform to a maximum of four operating systems (e.g., mainframe, Unix, Windows, and Linux) helps set plans for future rationalization of the portfolio. As Linux begins to mature for DBMS workloads (2005/06), volume Unix and Legacy OS workloads can move to Linux as the scaling and tools become available.
Myth #4: Server Virtualization Benefits
Due to the sheer performance of current platforms, much emphasis is on virtualizing workloads to increase platform use. On Intel x86 platforms, VMware/EMC and Microsoft with its Virtual Server offer tools to increase use. These tools are well established in test and development environments and increasingly are used in production. However, as these tools mature in capabilities and continue to gain acceptance, hardware will continue to decline in price. Therefore, the cost of the virtualization software license may be more than that of adding a physical blade server, in which case, from an asset and operations perspective, it is easier to provision a blade. Our research indicates that these virtualization tools integrate well with the ARM tools mentioned earlier. However, it then becomes difficult to assess the impact of the virtualization tool compared to the improvement in efficiency gained from deploying the ARM tools alone. We recommend deploying the ARM tools first and then virtualizing the hardware with blades and the software with these virtualization tools. This in turn increases systems administration ratios and use for workloads that require it.
Myth #5: You Can Displace Microsoft From Your Portfolio
Perhaps the biggest myth of all is that with server consolidation an organization can rationalize Microsoft out of the portfolio. Although an ITO may reduce some spending with Microsoft for file-and-print or infrastructure servers, if the ITO has .Net applications, has deployed Active Directory, or has mature SQL Server or Exchange implementations, Microsoft will not be displaced as a server OS. We believe this to be the situation for more than 95% of IT organizations. Legacy Microsoft platforms are excellent candidates for server consolidation, but consolidation should be to a new managed Microsoft platform before a migration to an alternative OS, since the tool references and environments will be more proven than Linux/open source alternatives.
Bottom Line: IT organizations need to be realistic about server vendor consolidation practices and expectations. They should throw away the vendors’ rose-tinted glasses and compare case studies, using an apples-to-apples comparison.
Business Impact: The biggest savings of all is attained by rationalizing portfolios, not by consolidating server platforms.
META Group originally published this article on 18 June 2004.