PMBusTakes Command of Data Center Power Issues

Apr 1, 2008 12:00 PM
By Brian Griffith, Server Power Delivery Architect, Intel, DuPont, Wash.

Solving data center power problems is an important issue for all manufacturers of servers today. System density (the growing server blades market, for example) and the cost of energy consumption are on everyone's mind. So, it is no surprise the development of products that solve these problems is paramount for any system manufacturer.

To get the highest-possible density and lowest energy consumption in any data center, it is critical to make the most use of all cooling and power-delivery resources. Cooling must be optimized by effectively managing airflow. System power consumption must be controlled to maximize performance without exceeding power-delivery systems. All this must be done while maintaining high availability. To support this demand, tools for monitoring and controlling power at the system, rack and data center levels are being increasingly introduced in today's servers.

PMBus is an open industry-standard communications protocol that is a key component in creating server features to help the data center manager address power issues. Understanding where and how much actual power is consumed at the data center, rack, system and subsystem levels can help optimize data center power-delivery and cooling systems. PMBus power supplies and voltage regulators play a major role in the power monitoring and optimization architecture.

The figure shows where these PMBus-enabled power converters are located in the systems throughout the data center. Baseboard PMBus voltage regulators (VRs) that are powering processors and memory offer the ability to monitor power at the subsystem level. System power supplies with PMBus provide system-, rack- and data center-monitoring capabilities.

New system capabilities allow users to control power consumption by limiting performance states of components, such as Advance Configuration and Power Interface (ACPI) performance states of processors. Processor performance states are used to make tradeoffs between performance and power consumption. Details can be found in the ACPI specification.^[1]

Power control capabilities, along with virtualization, can allow data center managers to optimize thermals by controlling hot spots, shifting work loads and maximizing the available resources of systems, power distribution units (PDUs), cooling units and uninterruptible power-supply (UPS) systems. PMBus power sensors located in the power supply, and on the motherboard, VRs are a key resource used in implementing these capabilities. Standardization around PMBus allows suppliers to focus on improving capabilities and cost, not on making new flavors of identical features.

PMBus commands for use in server power converters can be categorized into three areas: power sensing, thermal management and diagnostics.

Power Sensing

The PMBus commands critical for sensing power consumption in the system power converters are READ_PIN, READ_IIN, READ_POUT and READ_IOUT. The PAGE command in conjunction with the READ_IOUT command is used for converters with multiple outputs.

Each output of a power converter is assigned a PAGE number. Table 1 summarizes these commands and their associated system management bus (SMBus) protocols. SMBus is used in servers for low-speed communication. PMBus is based on the SMBus revision 2.0 specification.[2]

Want to use this article? Click here for options!
© 2007 Penton Media, Inc.