| More power to you |
| Thursday, 26 February 2009 00:00 | |||
|
Raritan says it is possible to improve uptime and capacity planning through rack-level power and environmental monitoring Organisations are more dependent than ever on computing resources to handle everything from mission-critical financial transaction applications to storing vast amounts of patient information. Data centre facilities are under increasing pressure to flawlessly add significant, reliable computing power whether the data centre is experiencing rapid growth, consolidating from multiple locations or trying to extract more capacity out of an existing space. While blade servers have addressed some of the challenges placed on IT staff, such as cramming more computational power into a limited amount of space, this new technology has also created its own set of problems. Today, both IT administrators and facilities managers are addressing questions like, “Am I getting enough power to the blades?” “Are there racks with excess capacity?” and “Am I getting enough cooling to those racks?” The fact is, HVAC systems are now being routinely pushed to their operational limits. The amount of power required at the rack is increasing dramatically – in some cases growing by an order of magnitude from 2kW to 20kW per rack. Plus, the sheer number of server units in a data centre is making it more difficult for limited IT staff to manage – much less improve – uptime. This article explores the challenges of increasing computing power in both an inflexible and resource-constrained physical space, as well as centres that are experiencing high growth. We also explore how companies can employ emerging support technology – like intelligent power distribution units (PDUs) – to help manage server uptime and capacity planning.
Over the past five years, the pressure to expand computing resources has led to an explosion in the demand for data centre servers. This, in turn, has driven the development of new technology, most notably blade servers. While it’s a welcome development for most IT professionals, because it means more resources can be shoehorned into an existing space, there is also a significant downside. As components, like server microprocessors, become more powerful, they consume more power and generate more heat. This has put a strain on the power and cooling infrastructure that supports them. In fact, according to HP, data centre power densities have grown from 2.1 kWh/rack in 1992 to 14 kWh/rack in 2006. This requires both facilities and IT operations to take a closer look at cooling systems and how they can better deal with local "hot spots" within the computer room. Heat and server failure Understanding how to move and cool air in a high density dynamic data centre is very important to maintaining uptime. Left unchecked, hot spots could lead – and have led – to unexpected server shutdowns and, over time, premature server failure. This, of course, results in unacceptable downtime, strain on your staff’s time and added pressure on your budget. Greening initiatives With the growing importance on the “greening of data centres” more executives are noticing the effect that power consumption is having on the world around them. They are witnessing • Increased emissions, e.g., greenhouse gases, from electric generation
There are a number of compelling benefits that can be achieved if companies increase their efforts to limit power consumption and improve power efficiency. Namely, management can look forward to: • Extending the life of existing data centres and delaying the huge costs associated with building new data centres Clearly, new technologies need to be deployed to help manage change, optimize data centre designs and plan for future growth. How will IT and facilities managers deal with change? Making a data centre more power-efficient takes more than a ‘one and done’ approach. It needs a commitment to design and execute a coordinated, long-term approach that focuses on efficiencies at both the technological and strategic levels. Some efficiencies can be found in the equipment itself. These can include powerful yet more efficient CPU chips, power supplies, servers, storage devices and networking equipment. There are also software strategies, such as server virtualization, that can play a significant role in reducing overall power consumption. While all good, there is also a clear need to do more: • To be able to “see” and manage the effects of growth and expansion, especially for capacity planning purposes Few solutions give you a better ability to monitor and control activity both at the server and rack level than the latest intelligent rack PDUs. These power distribution devices come equipped with features that measure power at the outlet level, as well as environmental conditions such as temperature and humidity at the rack. They also provide the IT professional with accurate, real-time data that he or she can then act on.
Today, PDUs come in a number of configurations that help you manage power in the data centre. This section will describe the four basic types and provide you with a summary of their strengths and weaknesses.
Basic PDUs are power strips that are constructed out of high-quality components for use in critical environments. They generally support distributing correct voltage and current to several outlets. Pros: Basic, affordable, proven technology.
Summary: Metered rack devices meter the power at the PDU level, e.g., RMS volts, current, etc., and display it locally. Typically, the meter displays information for the entire PDU. A few sophisticated models have user-defined alarming capabilities and the ability to remotely understand PDU-level metering data over a serial or network port. Pros: Provides real-time remote monitoring of connected loads. User-defined alarms alert IT staff of potential circuit overloads before they occur.
Summary: Switched PDUs provide controlled on/off switching of individual outlets and load metering (see metered PDUs above) at the PDU level. They allow users to power recycle devices remotely and apply delay for power sequencing and provide some outlet use management. Pros: Remote power on/off capabilities, outlet level switching and sequential power-up.
Summary: An intelligent rack power distribution unit can be controlled remotely via a Web browser or command line interface (CLI). It meters power at both the PDU level and the individual outlet level; supports alerts based on user-defined thresholds; provides security in the form of passwords, authentication, authorization and encryption; and incorporates rich environmental management capabilities. Models should also be highly customizable, support the industry’s latest standards-based techniques like SNMP TRAPs/SETs/GETs, IPMI and SMASH CLP, and integrate seamlessly to existing corporate infrastructures like LDAP, Active Directory, RADIUS and NFS servers.
So how do intelligent PDUs effect change in the data centre? Here are a number of situations, scenarios and opportunities to consider.
Idle or underutilized servers waste energy needlessly, so it is productive to identify, redeploy and/or decommission those servers as soon as possible.
Today, consultants are becoming increasingly sophisticated in designing and modeling new data centres. By applying complex software tools like Computational Fluid Dynamics (CFD), they can better plan cooling infrastructures and address existing data centres that have heat issues. However, these tools have their drawbacks if they are modeling based on nameplate data and static environmental conditions. In the real world, data centres are dynamic. Server utilization changes over time, which in turn causes power consumption and heat generation/dissipation to change, which in turn requires appropriate cooling to specific racks or rows of computers. Intelligent PDU technology can help provide information such as inlet temperatures at different locations within a rack over a period of time to aid in the planning of air movement and cooling. It can also provide alerts based on customisable thresholds so that hot spots can be immediately addressed to improve server uptime.
A common mistake is overcooling IT equipment in the data centre. While it is true that excess heat can shorten the life of IT equipment or even cause it to fail, excessive cooling does not prolong equipment life and is a waste of energy. In fact, according to ASHRAE, temperatures at the rack can now be as high as 77 degrees Fahrenheit as long as it remains relatively consistent in all locations. By placing intelligent PDU temperature sensors at the bottom, midpoint and top of a rack’s cold air inlet, not only can IT administrators determine if the system is overcooling the servers, but whether the cool air is being routed efficiently to each of those levels. Temperature sensors can also be placed at the rear, or hot air outlet of a rack to monitor the temperature of the exiting air. By measuring temperatures on both sides of the rack, IT administrators can get a more complete picture of the air flow in and around each server. This data can also be quite useful when servers, racks or cooling equipment are added or rearranged in the data centre, potentially leading to counterintuitive changes in temperature levels. For example, it is not uncommon for air conditioning units to be added to a server room only to discover that, due to altered air flow patterns, some racks receive very cold air while other racks are actually hotter than before the additional air conditioning.
Because intelligent PDUs offer both PDU and outlet-level metering, it is possible to accurately determine both the existing and potential power capacity at each rack and at each server. This information can help IT administrators make informed decisions when deploying new resources in the data centre, ensuring new servers and other IT devices are deployed where they won’t overload an existing circuit.
Server virtualization is a strategy that’s gaining strength in IT because it effectively reduces the number of devices required to run applications in the data centre. It is also an effective tool for capacity planning. However, virtualization introduces other concerns. Chief among them: with higher utilization rates, the physical servers that are virtualized are running at higher capacities, drawing more power and generating more heat. Additionally, if multiple high computational lead applications are running on a server at the same time, it is conceivable that they may reach an operating threshold that threatens to shut the server down. By being able to measure the outlet-level power consumption of these physical servers over time, IT administrators can deploy applications so they will not exceed set levels, e.g., 70 percent of nameplate capacity.
With intelligent PDUs leading the way, data centre staff will be able to more accurately identify where and when servers are running too hot or too cool, save energy and avoid server crashes; identify, redeploy or decommission servers, to balance power usage more uniformly; and identify trends and opportunities to more intelligently manage change and growth. In other words, intelligent PDUs can help you know more about your data centre so you can better manage your IT operations and resources.
|
