Vendor hype is prevalent regarding grid computing, server consolidation, virtual machines, and server blades. Some of these concepts have been tried and tested, while others are still only vendor visions and road maps. IT organizations (ITOs) must be pragmatic about these programs and either balance the risk of being bleeding edge/leading edge or wait until mass adoption occurs.
META Trend: Business brand image will depend on users' quality of experience, with infrastructure scalability and availability being critical prerequisites. Robust infrastructure will be implemented via a multi-tier mix of small Web and application servers deployed increasingly on server blades that will address the density issue and force use of a single point of management on all servers, not just blade form factors (2003/04). Database scale-out clusters that leverage a networked storage (SAN) tier will offer good-enough performance for scale-out clusters on Unix (2003/04), Windows (2004/05), and Linux (2005/06), comparable to that of big SMP.
Through 2008, hardware and software virtualization will replace server consolidation as a principal IT program. Moore’s Law (the doubling of price/performance every 18 months) is driving commoditized Intel server platforms to replace proprietary Unix/RISC servers. The hardware itself is less complex, but the workload is not - it requires better virtualization and server management software. Through 2005, Intel server virtualization will leverage modular (blade) servers, necessitating increased use of infrastructure management tools.
We are now seeing vendors and ITOs deliver improvements in workload scheduling and server capacity, which is leading to increased overall use. Virtual machine (VM) tools such as those from VMware now offer virtualization techniques for commodity Intel server platforms that were long ago introduced on the mainframe. Through 2007, Intel server use will peak at 80% (increasing from 10%-60% currently) through use of VM capabilities. This growth will be capped by the limited I/O scalability of Intel hardware. Since these VM platforms are not as mature as a native mainframe, which is capable of scaling to 95% workload, significant changes will need to be made to current Intel architecture for this type of workload to be achieved.
Through 2006, commercial grids are not expected to have considerable impact on an ITO’s portfolio. Technical grid technology is maturing, as highlighted by the successful implementations of technical Linux clusters, yet commercial grids will be severely hindered by lack of standards and tools for management, measurement, and usage tracking. Immature Web services and lack of flexible software licensing, along with inadequate usage models, will inhibit grid adoption. Many high-end Unix platforms now offer basic measurement, management, and virtualization extensions to their proprietary environments. These can be seen as a “grid-in-a-box” - the 21st century rebirth of the “LAN-in-a-can” - but the simple fact that they are co-located on one high-end platform does not protect them from the same obstacles of server grids. This begs the question: “When is a mainframe not a mainframe?” The answer: “When it is a Unix mainframe.”
Our recent research tracks the adoption of consolidation, grid computing, virtual machine tools, and blade servers, based on feedback from 25 end-user clients in three European virtualization and consolidation workshops. The informative feedback from these workshop sessions, in a pragmatic sense, highlights the true status of virtualization programs in the data center. The outcome breaks down IT projects covering virtualization and categorizes the projects as follows:
Grid and utility computing concepts have been touted by the likes of IBM and HP for three to four years, and more recently by CA and Oracle. Unlike server consolidation, grid technology is just maturing, with commercial deployments for grid computing almost non-existent. Only in pharmaceuticals and in the chemical industry, which have high research and development budgets, do we see a commercial crossover of technical computing clusters into grid-based initiatives. Through 2007, software licensing will continue to be a big issue for grid acceptance. This will be further exposed and addressed with the maturing of Web services for production platforms. ITOs should also evaluate grid concepts in the testing and development environment.
Virtual Machines and VMware
Virtual machine capabilities have been available on mainframes for many years. VM concepts have been introduced for Intel platforms by companies such as VMware and Connectix (which was acquired by Microsoft). These VM platforms have brought vast improvements to the virtualization and consolidation capabilities of Intel platforms. We have been positive about VM on Intel and are now seeing commercial maturity and improving technical capabilities in these products. Many clients are considering VM capabilities, but many see the attraction primarily in the testing and development environment. The lack of I/O scaling of the VM engine results in these tools being applicable only for certain workloads. Processor-centric workloads are good candidates for VM consolidation; I/O-centric workloads can expose I/O constraints of either the host VM software or the lack of I/O scaling of the native hardware platform. We expect these I/O issues to be addressed during 1H04 with maturing VM SMP support.
Our research indicates that ITOs are also concerned about the size of companies offering these products. VM on the mainframe is from IBM, while on Intel, it is from VMware (a medium-sized company) or Connectix (in the process of being digested by Microsoft). This is a main reason why ITOs partner with hardware suppliers for VMware support and contain the VM products in the testing/development stage. However, the biggest threat for VM on 4-way Intel platforms is the use of blades servers with hardware management software (e.g., IBM Director, HP Insight Manager/Service Guard). Through 2007, the real attraction of Intel VM platforms will be on blades servers, partitioning each blade as multiple VMs. The technology of scaling across blades or servers will not mature until 2008.
Blade form factors have (surprisingly) been accepted in the IT portfolio (see SIS Deltas 979 and 982). More than half of the ITOs in the study are considering blade platforms. Very few have them in testing or development and many have them in production. When asked, ITOs stated that the testing/development phase is minimized by use of infrastructure management tools and the hardware management software. Therefore, the blade servers easily “slip” into production on the back of using such tools. When asked further, the ITOs with the blade servers in production had previously invested in management frameworks. The ITOs that were considering blades did not have mature usage of these tools. Blade adoption is made easier by investing in the tools and frameworks upfront, rather than using tools and form factors at the same time. Indeed, the real attraction of blades is not form factors, but management.
We expect these end-user consolidation workshops to continue through 2004, which will give us the opportunity to reiterate these technology adoption observations and recommendations.
Business Impact: New virtualization technologies require more investments for testing, development, and integration, compared to traditional platforms.
Bottom Line: When introducing new server platforms into production for consolidation, grids, blades, and virtual machines, ITOs must be pragmatic about the vendors' marketing promises, and be certain to examine their references and service/support capabilities.
META Group originally published this article on 2 October 2003.