As server virtualization software evolves, so too does the hardware on which it runs. Keeping up with hardware changes as they relate to the software is always a challenge. This column will give you a quick look at the technologies and trends in processors and servers, relative to virtualization, blades versus racks and Intel versus AMD.
When x86 server virtualization emerged around 2001, it required a lot of workarounds, because of the x86 architecture. VMware made a significant investment to implement virtualization on hardware that was not well suited for it. Over the past few years, both Intel and AMD have implemented virtualization hardware assists that both enable and optimize virtualization on their chips - Intel-VT and AMD Pacifica. As a result, other virtualization alternatives such as Xen 3.0 and its derivatives now operate on both these processor lines. Although these virtualization assist chips are not required for VMware ESX, it does utilize some functionality today (used for improved performance for 64-bit guests), and will likely leverage additional capabilities going forward. These chips are now a standard option for x86 servers, both blades and rack-mounted.
Additionally, Intel plans to expand their technology to the I/O subsystem with a feature called VT-d, bringing virtualization to a new level for interfacing with networks and storage. This is an example of a much larger trend - moving towards virtual I/O in a variety of ways. Variations of virtual I/O capabilities have appeared within blade server systems for the past 5 years, with virtual I/O pooling and diskless/stateless bladed pioneered by Egenera. HP and IBM are now pushing hard in this area as well, for example with HP's Virtual Connect. Blade systems currently are leading the way in this area.
Of course, virtualization has also had a huge impact on the server business in general. Since virtualization allows the consolidation of multiple virtual servers onto one physical server, it reduces the number of physical servers required to accomplish the same workload (exact ratios are of course dependent on the workload and size of the server). For vendors and channel partners, this can translate to reduced revenue in server sales. For those sales organizations that resisted virtualization as a result, the short term effect may have been to postpone virtualization for a time, and keep server revenue up. However, the long term effect, as users began to see the benefits, (many are just now beginning to understand this) is that users started taking a new look at their partner relationships. Many gave new consideration to alternative hardware partners who were also knowledgeable of, and selling virtualization software options, either bundled or unbundled. The message here is clear, as server virtualization continues to take hold, it will have an impact on server sales, and shift the revenue model more from hardware to software and services. Those who are prepared, and shift their business accordingly will fare much better.
Blade servers vs rack servers
As space, power and cooling continue to loom as challenges to IT, blades are becoming more attractive as a consolidation platform. The blade market has become a $3 billion dollar a year business, with significant growth over the past year. Server blade revenue was up 18% in the last quarter of 2006, and up 30% for the year. By comparison, overall server revenue was up only 2% for 2006. However, the total revenue is still only 5.2% of the overall server market.
Interestingly, the adoption rate for blades seems to be following the same curve as the adoption rate that occurred with rack servers, based on the number of years from introduction of each technology. Forecasts estimate that between 20% and 30% of servers will be blades by 2011.
It's also interesting to note that market share numbers in server blades shifted in the fourth quarter 2006, with HP taking the lead away from IBM. Fourth quarter numbers show HP now having 42% of the market and IBM having 37%. However, IBM remained in front for the year, with IBM holding 40% and HP holding 37%.
In the early days of blades, one of the main objections was that blades were limited in their options, in terms of processors, memory and I/O capability (e.g., number of NICs, number of HBAs allowed per blade). This limitations have been overcome in blades today, and the sytems vendors have now all committed to making the same options available on blades as they deliver on rack-mounted servers (e.g., HP calls this their "Blade Everything" strategy). With high-end and multi-core processors on blades, consolidating with server virtualization on blades offers a double whammy in reducing space, power and cooling, with both technologies working together to shrink the footprint and the costs to perform the same workload.
Another example of this shift in blades can be seen in Sun's recent server strategy. Their first Intel based server ever is being delivered on a blade, rather than a rack, later this year. It will then be followed by two- and four-socket rack versions. Sun had originally entered the blade market early on, and then withdrew, and reentered last year with a completely edone, strong offering, but late in the game.
AMD vs Intel
The x86 market has been an interesting two horse race. After a long held lead, Intel began to lose both market leadership perception and market share, when AMD took the lead with its Opteron chip. Intel showed an 11% drop in revenue in 2006, while AMD almost doubled sales. More recently, though, Intel has come back into the game by beating AMD to market with its Clovertown quad core Xeon chip, which is shipping now. The new 5300 quad-core is 50 watt max (compared to 80 previously), with 8 MB of on-die cache. Estimates have been given of a 50% Performance improvement over the previous Woodcrest chip. However, some argue that the Intel quad-core as delivered is really just 2 dual-core Woodcrest chips glued together. AMD has announced its Barcelona quad-core Opteron, to ship later this year, describing their approach to quad core as "monolithic" or "native" quad-core. Barcelona has four cores on a single piece of silicon. Of course, according to AMD, Barcelona will improve both performance and efficiency over both Opteron and its competition. Benchmarks will really be the only way to tell.
Servers based on the new Intel products are expected to be available worldwide over the next few months from Acer, Dell, Digital Henge, Fujitsu Siemens, Hewlett-Packard, HCL, IBM, Rackable Systems, Samsung, Verari, Wipro, and other companies. AMD expects server vendors such as Dell, Hewlett-Packard, IBM, and Sun Microsystems to adopt its new technology for their servers in 2007.
Intel also led in converting from a 90 nm to a 65 nm process. AMD is now moving to a 65 nanometer process with Barcelona. Intel is moving to a new 45-nonmeter manufacturing process late 2007, which Intel said it would ship in the second half of the year. Code named Penryn, the new processor line is believed to cross product lines from desktop to server, workstation and mobile in the long term (with both dual-core and quad-core chips). The goal and benefit of going to a smaller process, i.e., the 45nm process, is that it builds a smaller chip with less power leakage than the 65nm and 90nm predecessors.
Each vendor seems to be taking a turn in the winner's box for varying features. In one niche area for example, the high-end of system fault tolerance, NEC has just announced their newest generation of servers that provide Fault Tolerant lockstep support of the Xeon, the first to support quad core CPUs. (This type of system fault tolerance provides a major step up from what most consider high availability.) The new NEC FT server uses the new Cloverton quad core chip from Intel. According to NEC, they couldn't run an AMD chip in lockstep and so they based their new fault tolerant server on the Intel chip. Other differences (including the memory controller not being in the processor and the front side bus used by Intel vs hypertransport with AMD) were part of NEC's processor decision as well. (NEC does utilize AMD chips for other offerings).
The race continues with competing future plans in various areas. While continuing to compete on specific features, benchmarks can be created to allow either side to win. The question is who wins in your environment with your workload and when considering performance per watt.
For more information on server/processor issues and trends related to multi-core chips, power issues (going green) and licensing, check out my related column Technology trends in processors and servers on SearchSystemsChannel.com.
Blades and Virtualization Summit
For more in-depth, up-to-date information on server chip issues, power and cooling improvements, futures in virtualization, blade server implementation, and more, consider the Blades and Virtualization Summit on May 1-3. You'll hear from and connect with blade server hardware and virtualization software vendors, power and cooling companies, TechTarget expert panelists, and other experts on the latest products and services related to these technologies.
About the author: Barb Goldworm is president of Focus Consulting, a research, analyst and consulting firm focused on systems and storage. She has spent 30 years in technical, marketing, industry analyst and senior management positions with IBM, Novell, StorageTek, Enterprise Management Associates, and multiple successful startups. Barb has authored numerous business and technical white papers and articles and has just finished a book, Blades Servers and Virtualization: Transforming Enterprise Computing While Cutting Costs.