The domination of Windows 2000 as a server operating system and the rise of Linux in the data center (unifying Unix as a Microsoft alternative) are clearly benefiting one player - Intel. As platforms are commoditized, IT organizations must drill down and ratify their selection criteria for Intel Server Platforms.
META Trend: With distributed n-tier (DBMS, application, Web) server architectures standardizing on Intel, proprietary Unix (Solaris, HP-UX, AIX) will recede to high-end, low-unit-volume, legacy-platform status by 2005/06, displaced by OSs designed for Intel economics: Windows and Linux. Linux will rapidly mature and gain momentum as an ISV reference platform, moving beyond high-volume Web, technical computing, and appliance server environments into mainstream application and DBMS server roles by 2004/05. Linux server growth will initially be at the expense of Unix (2003/04), but will eventually vie for dominance with Windows (2005/06).
By 2007, Intel platforms will dominate the data center with a majority presence. By 2012, this will rise to more than 80% Intel, 17% proprietory or RISC, and 3% legagy CISC (mainframe). This growth will be fueled by the sustained migration from Unix to Linux platforms and the entrenchment of Microsoft Windows moving from a “good enough” operating system to a mature platform fueled by .Net adoption. Intel will clearly win the battles in the data center war in volume (units) and value (dollars). However, Intel’s worst competitor will be itself. Ever-fragmenting and forking processor road maps are causing confusion and overlaps in the postioning of Itanium (2 and LV) with Xeon (DP and MP). Intel’s hardware OEMs and software partners (e.g., Microsoft) are now maintaining many aspects of the Intel offering.
Through 2005, 95% of all workloads will be addressed by Intel platforms. We believe that 80% of these platforms will have a 32-bit Xeon base. Itanium platforms will need to gain enough momentum to garner more than 20% of the market before Itanium will be accepted for the potential volume market and gain more aggresive support from Microsoft. Until that time, it will remain primarily a Linux or Microsoft SQL Server play.
Our research indicates that IT organizations (ITOs) must evaluate Intel Server Platform vendors using an additional set of metrics and criteria. This will help to measure and position the hardware vendor as tactical (in price), strategic (across the portfolio), or leveragble (with service and integration). These criteria should be used alongside the tradtional metrics for all server platforms. Unifying Linux/J2EE and Microsoft .Net infrastructure platforms helps standerdize and reduce costs across the portfolio. ITOs should push Intel Server vendors with Windows and Linux substitutability.
As platforms become more modular, the occurrence of a server’s operating system may migrate back and forth from Unix to Windows or Linux. This change must be controlled as part of a managed framework. Our research indicates that a balance between price/performance and support/management helps to establish the maturity of Intel platforms in the ITO’s portfolio. Key metrics that ITOs should assess for Intel platform vendors include:
- Web servers (1- to 2-way): The vendor should offer conventional high-density, rack-mounted servers (1U/2U) for Web and edge functionality. Focus should be given to I/O, network, and storage capabilities, depending on the scale-out workload requirements and operating system capabilities. Little differentiation exists between the capabilities of the Web server platforms. Traditionally, IA-32 Xeon platforms address this space.
- Application servers (2- to 4-way): Server vendors can still differentiate themselves in the application server space, but the standard 4-way is available as a white-box platform with supporting chipset from Intel (through its OEMs). There are negligible differences in the current production platforms. Focus should be given to I/O capabilities and the additional support of increased memory as required for midtier application servers. Management of multiple application servers is key in using adaptive resource management (ARM) tools. IA-32 Xeon MP platforms comfortably address this space in a Lintel or Wintel stack.
- DBMS servers (8-way+): High-end database servers still require heavy R&D budgets from hardware vendors. The 8-way capabilities are equaled by next year’s 4-way, narrowing the return on investment for budgets. ITOs should be careful in their deployment of any platform that is more than 4-way, since price premiums may be avoided by deploying on smaller platforms where possible. Likewise, the high-end consolidation promises of vendors are often oversold and underachieve in production environments. Xeon MP still addresses 90%+ of workloads in the DBMS space, but Itanium platforms are now gaining ground in large SQL Server deployments.
- Blades and bricks: As Web and application servers move to commoditized infrastructure platforms, new form factors such as brick and blade servers become attractive (for consolidating large server farms into smaller areas and reducing power consumption). Server density and the use of ARM tools can double server system administration ratios (from 18 to 30+ per full-time employee). Yet ITOs need to be cautious regarding heating, cooling, and power capabilities of blade servers and the limited I/O capabilities of blades. Through 2005, we expect blade servers to benefit most from Centrino mobile processor technology and PCI Express I/O.
- Itanium offerings: Itanium platforms currently are limited to DBMS workloads, technical computing, and HPC with Linux. Through 2005, Itanium will gain acceptance for midtier application server platforms, and its significance for the vendor’s portfolio will increase. Itanium is on the brink of being successful - again.
- Management tools: As with any group of servers, the management framework for the platform is critical. Deploying ARM tools such as HP’s Insight Manager or IBM’s Director helps migration to Linux servers (from Unix) as well as management of Microsoft coexistence within a portfolio. These ARM suites must be consistent across blade, Web, application, and DBMS servers.
- Price: Clearly, the lowest price is preferred, but ITOs need to compare street price (postdiscount), and be consistent.
- Midrange storage: As Windows 2003 and Linux help drive externalized/network storage for all server platforms, provision of midrange storage, fabric, and software is crucial. This helps the ITO deliver a mature set of storage services for all server tiers and types as well as address potential I/O issues across the stack.
After gaining understanding of the relevant metrics, an ITO can then assess the capabilities of the Intel Server Platform vendor. Our research indicates a diverse range in vendor performance across all these metrics, with no clear leader in every metric (see Figure 1
). We position the server vendors for Linux platforms as follows:
Business Impact: Deploying adaptive resource management tools can increase server system administration ratios from 18 to 30 per full-time employee, significantly reducing operational costs.
- HP: HP does well across the board, but it still is significantly higher priced than Dell (by approximately 30%), though it will discount to within 10% when pushed. At the high end, IBM positions the x440 above HP’s IA-32, but we expect HP’s Itanium efforts to balance this out through 2004.
- IBM: IBM has made great strides in the high-end Intel and blades space with IBM Director, yet it is weak in the Web and application server market. We believe the promotion of IBM BladeCenter is helping it to overcome its weakness in this space. ITOs should consider blades for density and management capabilities.
- Dell: Dell’s strength of price and value is an important aspect of its success in the data center. Its weaknesses include poor high-end positioning, lack of a blade strategy, and the poor acceptance of its OpenManage as an ARM. However, we believe Dell will overcome these challenges by partnering and improving on its services capabilities, similar to its partnering with EMC for midrange storage.
- FSC: Fujitsu does well across the board, but needs to reproduce its successes consistently worldwide.
Unisys: Unisys has been integral to Microsoft and Intel for high-end messages with its CMP platforms scaling to 32-way. Its recent modular servers have more attractive price entry points, but HP and IBM are becoming increasingly more difficult for Unisys to displace. Unisys and Dell complement each other, and reseller agreements now exist between them that will help both companies overcome the gaps in their portfolios.
- Sun: Sun has only just entered the Intel Server space with its initial products. Only pure Sun shops should consider use of Sun’s offerings for Intel deployments, and even then, price is high and management remains weak.
Bottom Line: When selecting an Intel portfolio, a balanced scorecard is required to reflect the breadth and depth of the vendors' offerings.
META Group originally published this article on 27 October 2003.