|
Zero DownTime Magzine
– July 2008
The world of high density applications and computing has progressed rapidly in the past ten years. Traditionally dominated by large financial and industrial companies who required robust parallel processing to perform a huge array of complex business functions, high density computing has become the standard for virtually all enterprise business, and an increasing proportion of medium-sized companies.
Two worlds converging: high availability and high density
Historically, not all high availability applications required a high density computing environment. Also not all high density applications required a high availability platform. In its earliest form, dense computing was used only by seismic processing and reservoir modeling applications requiring an N+1 power infrastructure, at best. Although these systems required more power and cooling, it was acceptable for them to be restarted in the event they overheated or in the event power was distributed.
Because high density hardware platforms dramatically improve business performance, they have become the de facto standard for next generation back office application design, thereby converging the high density and high availability worlds into a necessary datacenter solution. The business world as a whole has shifted to a new paradigm where all back office applications reside on higher density systems. For example, today’s global supply chain management applications – ubiquitous in manufacturing and distribution industries – can achieve optimal performance only when hosted on dense computing architectures.
The concept of high availability/high density computing is engineered specifically to accommodate “always on” applications that require robust processor speeds and vast data storage capabilities. In many enterprise-scale ERP models, applications must be deployed with maximum (ie: permanent) availability for thousands of users across multinational geographies. Only carefully configured high density computing and datacenter environments can deliver this level of global availability and productivity.
These requirements for high availability computing have driven a major shift in datacenter architecture from an N+1 model to a 2N configuration. A typical enterprise-scale configuration previously might have had several power components such as UPSs and generators with one dedicated backup unit (N+1). Because of the increased requirements for high availability, datacenters now have to install a replicated power model that would require a separate UPS or generator for each UPS or generator in service (2N). Since those servers would typically run at 30-40 percent capacity and could be easily partitioned, having a single backup unit for multiple servers was sufficient. With high density capabilities, those eight servers have been condensed into a single robust server with a single dedicated backup server (2N). While this improves system backup and helps ensure business continuity, high-density servers host a far greater quantity of data and applications than previous server generations. They run constantly at maximum or near maximum capacity, meaning the hardware is constantly operating at higher processing speeds and at an elevated temperature. This requires greater temperature control and environmental monitoring.
These changes have also been driven to a large degree by server OEMs. Research and development and product lifecycles architectured by hardware manufacturers are in many ways forcing the migration to high-density/high availability computing. This migration, though, offers remarkable potential business benefits. The music industry’s transition from cassette tapes to compact discs in the late 1980s provides a good analogy. Customers and manufacturers alike were excited about the benefits of using CDs, which offered far greater possibilities than cassettes. Music industry executives recognized that CDs were the platform of the future. No one foolishly stood their ground and proclaimed, ‘We’re going to continue manufacturing cassettes because that’s how it’s been done for years. It works for us, so we think it will continue working for everyone else.’ In order to stay competitive and remain relevant in the market, you must adapt and adopt – adapt to market changes, and adopt new platforms.
Virtualization drives changes
High density computing environments have also contributed greatly to the rise in datacenter virtualization, though it could be said that the relationship is reciprocal – the demand for virtualization increases the need for well-architected high density datacenters. Virtualization within a centrally-located server bank provides a means of managing systems and resources functionally regardless of physical layout or location. It’s a relatively new business continuity technology that augments “always on” availability and supports growth in missioncritical, high availability applications.
For example, a company before virtualization may have run each of their business applications on separate servers. Since those servers would typically run at 10-20 percent capacity they would produce a relatively small heat output. With virtualization, those same servers have been condensed into a set of small robust servers. While this improves system performance, the servers are now constantly at maximum or near maximum capacity, meaning the hardware is constantly operating at higher processing speeds and at an elevated temperature. This requires greater temperature control and environmental monitoring that can only be delivered by high density datacenters.
According to Forrester Research, more than onethird of enterprise businesses have implemented server virtualization. This trend is expected to continue, with 75 percent of enterprises deploying some form of server or application virtualization by 2010. North American companies lead the global shift toward virtualization, with EU firms a close second.
Competition continues to grow among virtualization platform providers as they ink agreements with hardware OEMs such as HP, Dell and IBM to deliver out-of-the-box virtualization solutions. In fact, this market is expected to exceed $7 billion by 2011 – up from $800 million in 2006. From a datacenter perspective, this means we can expect to see a complete systemic transition to high density computing infrastructures and protocols within the next several years.
Impact to the data center
High density computing requires a unique IT platform, one that has had a dramatic impact on datacenter design. High density/high availability applications are consolidated, hosted and managed on blade servers, which require less physical space than traditional servers and can be condensed into ultra high-performance racks that deliver unprecedented performance. These modular racks take up little space and are relatively easy to update or reconfigure.
Blade servers are designed for maximum processing ability, and this increased performance comes at a hefty cost, literally – blade servers consume, typically, 10 times more total power than conventional rack servers, and require at least a four-fold increase in cooling capacity. As a result, according to a recent Uptime Institute study, power consumption in datacenters has increased sevenfold (over 600 percent) in just seven years.
High density blade servers operate at 50 amps, a major increase over previous server generations that typically ran at 20 amps. As a general rule of thumb it takes $1 to cool every dollar in direct consumption in a datacenter. For example, a 250kW consumption rate would equal to $43,200 a month in utility cost or approximately $2,600,000 over a five year period.
In fact, companies today are confronting dramatic cost increases across the board in terms of building and operating a high density, high availability infrastructure. According to The Uptime Institute, a Tier IV facility with 25,000 square feet of raised white floor at 100 watts per square foot can cost up to $60.5 million. High density server technologies require 200+ watts per square foot, scaling these costs even further. Beyond the rising building costs, capacity planning is also a challenge for companies. Overbuilding an infrastructure, where white floor space is underutilized, can be detrimental to the return on the investment. Under-building or poorly retrofitting an infrastructure that is obsolete and doesn’t meet user-demand creates additional problems.
In addition to consuming greater power, high density environments also generate more heat. This places greater emphasis on air flow and temperature controls. High density datacenters typically use a computational fluid dynamics, or CFD, simulation to engage the entire layout of the datacenter and run through all the “what if” scenarios to identify hot spots and potential problem areas. The sub-floor ventilation and perforated tiles are designed and laid out using engineering-driven schematics that maximize output and availability for any emergency overheating situation. A dual plenum design allows for a hot aisle/cold aisle configuration to effectively disperse and control excess heat, making emergency measures immediately effective at cooling down the problem area, whether it’s an entire zone or an individual rack. CFD simulations are also highly effective at identifying the strategic placement of CRAC units to ensure that optimal airflow can be diverted and/or increased to any hot zone at any given time.
Fault tolerances are also identified concurrent with CFD simulations. Built-in recovery times are evaluated and rigorously tested – because they run high availability applications, dense computing servers require absolute minimal recovery times with zero interruption of availability. Recovery capabilities are ensured via power redundancy, which typically include primary-tobattery protocols ensuring business continuity and zero downtime while the center switches to backup generators. This also has to be true on the mechanical side of the datacenter. The use of redundant chillers and thermal storage are becoming necessary to assure that proper temperature and air flow are maintained.
While virtually all high-density server OEMs incorporate fault tolerance elements into the hardware design, it is essential to identify fault tolerance beyond the individual server or rack to include the impact of overheated hardware on an entire row or room. Hot-swapping malfunctioning components, while essential and integral to blade server design, is confined to dedicated servers and/or racks and typically does not include protocols for large-scale overheating and temperature fluctuations.
All of these power, infrastructure and environmental factors translate into three or four times the total cost of ownership for the life of the asset. Power cost in particular remains a daunting challenge for IT administrators – harnessing and conducting sufficient electrical and cooling power in times of uncertain energy costs can blow a budget out of the water in short time. This in turn creates a scramble to minimize the fallout on operating expenses. Some of the more progressive datacenters are already turning to reflective external building paint and automated misting systems to help control ambient room temperatures and further minimize electric and cooling expenses.
The outlook for high density computing
According to IDC research data, approximately 75 percent of medium-to-large enterprise companies are in the process of implementing high density computing platforms. By 2012, this number will approach 100 percent. While infrastructure and design challenges can be anticipated with near certainty, power and cooling requirements cannot. The principal challenge moving forward is to develop datacenter models that can mitigate these increasing costs. IDC anticipates that annual cooling and power costs could exceed 70 percent of high density computing expenses, far outweighing the purchase cost of hardware and components. This places an even greater premium on datacenter design that optimizes power and cooling distribution with maximum utility efficiency. High density hardware is essential to running high density applications, thus these costs are unavoidable and must be approached with new protocols to minimize the impact on IT budgets. These changes are daunting, but also exciting. The world of opportunities available through high density applications is just beginning to emerge. Many enterprise companies are currently achieving strategic and competitive advantages previously unimagined through the use of high density platforms. As the playing field levels and high density computing becomes ubiquitous, competitive advantages will be realized by those with the most reliable and efficient datacenter architectures in place.
|
NEWS,
EVENTS & 
