PMBusTakes Command of Data Center Power Issues

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

READ_PIN for AC-DC Supplies

The capability of reading the ac input power to the power supply (which is also the ac input power to the system) is the most important PMBus sensor to the data center manager and is used in many of the new rack-mounted systems for power management. The ac input power can be used as described in the following three READ_PIN usage models:

• Control of system power

  By monitoring the total system power, setting limits that the system will not exceed and using the system's power-control capabilities, the data center manager can guarantee that the system, rack and data center power will not exceed a predefined threshold. This protects data centers from developing hot spots.

• Optimization of power delivery and cooling equipment

  By monitoring the power consumed by each system, the data center manager can better manage the other resources in the data center to make them more efficient, like computer-room air conditioners used to cool rows of racks and UPSs/PDUs used to deliver power to racks. This helps reduce the cooling system's energy consumption and improve system density.

• Charge back

  In co-located environments, data center managers may want to bill end users by the system energy consumption. The ac input power monitoring can be used for this billing.

In all these cases, the accuracy of the input power (watts) is critical to ensure there is limited performance impact, reliable protection and proper billing. Good accuracy of the power supply's ac input wattage sensor is a valuable feature to the data center manager.

Calculating the input wattage to the ac-dc power supply is pretty simple. Multiply instantaneous voltage and instantaneous current on the ac wave shapes, then average these values over the needed period. This results in an average of the real input power (watts) to the power supply. The different averaging methods are described in the previous Averaging and Accuracy section on page 15.

READ_IIN, READ_POUT and READ_IOUT

READ_IIN is used by the system to monitor loading levels on PDUs and associated circuits. Rack power is sometimes limited by the circuits powering the rack. By monitoring rms input current, the data center manager can maximize the number of systems in the rack while still protecting against overload conditions on the circuits. This helps to increase system density in the data center.

Calculating the rms input current (and rms input voltage) can be done in two ways:

• True rms calculation

  By saving the data points over at least one cycle of the ac input, the rms current can be calculated by

  The disadvantage of this method is that more difficult calculations may require a more-expensive primary-side microcontroller. The advantage is accuracy; it generates an accurate rms result independent of distortion on the input current.

• 1.11 × rectified average

  If the input wave shape is a true sine wave, then the ratio of the rms value to the rectified average value equals 1.11. Using this ratio, a simple average value can be calculated and multiplied by 1.11 to generate an rms value. This is a simpler calculation than a true rms calculation. However, with a distorted input current, errors will be introduced.

The READ_POUT and READ_IOUT commands can be used by the system in a few ways. The resulting data can be used to determine percent loading status of the system power supplies and motherboard converters. This will let users know whether they have power available to fit more memory or an additional adapter card.

Another use is to tell the power-management agent in the system how much power is going to different parts of the system. The power converters that power memory will tell users the total memory power consumption. The converters that power processors will tell users the processor power consumption. This allows performance/power optimizations by the system power-management agent.

Side-Band Interrupt

Since the typical server system is polling the power converters at such a slow rate, there needs to be a side-band interrupt method if a quick response is needed by the system to react to a condition in the power converter. The SMBAlert# signal is used for this.

The protocol details of the SMBAlert# signal are described in the SMBus specification.^[2] A power converter asserts SMBAlert# by pulling it low after one of the limit thresholds has been exceeded. The system responds by sending the Alert Response Address to all the power converters on that SMBus branch.

The power converter that asserted the SMBAlert# line responds with its address and releases the SMBAlert# line. Table 3 lists some of the SMBus commands that can be set to assert SMBAlert#.