|C H A P T E R 4|
Server Power and Cooling Requirements
This chapter provides information about important power issues relating to your servers. Your server documentation provides more detailed power information.
The design of your electrical power system must ensure that adequate, high-quality power is provided to each server and all peripherals at all times. Power system failures can result in server shutdown and possible loss of data. Further, computer equipment that is subject to repeated power interruptions or fluctuations experiences a higher component failure rate than equipment that has a stable power source.
Topics in this chapter include:
Each server, when properly configured and installed, must receive sufficient incoming power to supply all installed components. The data center should be able to provide a stable, dual-current path to the installed equipment. In addition, the power infrastructure must be designed to maintain system uptime even during disruption of the main power source. It is important to use dedicated breaker panels for all power circuits that supply power to your servers. The power system should be designed to provide sufficient redundancy, eliminate all single points of failure, and allow the isolation of a single server for testing or maintenance without affecting the power supplied to other servers.
It is important to secure multiple sources of power when possible. Ideally, multiple utility feeds should be provided from different substations or power grids. This setup provides power redundancy and backup.
The servers provide power input fault tolerance via redundant power supplies. Therefore, it is prudent to attach to each primary power supply a common power cord from one power grid that can supply power to all servers, and to attach another power cord from a different power grid to the redundant supplies. If a primary power grid goes offline, a backup power grid will provide power to the redundant supplies to keep the servers operating. See Power Supplies for information about power supply redundancy.
Using an online uninterruptible power supply (UPS) and a backup power generator provides a good strategy for obtaining an uninterruptible source of power. The online UPS filters, conditions, and regulates the power. It protects the servers from fluctuating voltages, surges and spikes, and noise that might be on the power line. The battery backup for the UPS should be capable of maintaining the critical load of the data center for a minimum of 15 minutes during a power failure. This is typically sufficient time to transfer power to an alternate feed or to the power generator.
The backup power generator should be able to carry the load of both the computer equipment and the supporting heat, ventilation, and air conditioning (HVAC) equipment. The generator should include dual power distribution switch gear with automatic transfer switching. To offset the possibility of a generator failure, power system designers often include a temporary generator for secondary backup.
Grounding design must address both the electrical service and the installed equipment. A properly designed grounding system should have as low an impedance as is practically achievable for proper operation of electronic devices as well as for safety. It is important to use a continuous, dedicated ground for the entire power system to avoid a ground differential between various grounds. Grounding design in the United States should comply with Article 250 of the U.S. National Electrical Code unless superseded by local codes. Use an antistatic wrist strap when working inside the chassis.
All properly installed Sun servers are grounded through the power cable. However, there are reasons for installing an additional mechanism to equalize potential. Problematic or deficient conduits can negatively affect another server, especially with respect to the possibility of spreading voltages. Additional grounding points help to avoid leakage current, which prevent system malfunctions. Therefore, additional cables might be used to connect Sun servers and cabinets to the data center's potential equalization rail. Enlist the aid of a qualified electrician to install grounding cables.
A primary power switch that can disconnect all electronic equipment in the data center is specified by NFPA 70 and NFPA 75 (National Fire Protection Association specifications) at each point of entry to the data center. The primary switch should disconnect power to all servers and related electronic equipment, HVAC equipment, UPS, and batteries. Multiple disconnects for separate parts of the power systems are also acceptable, but in both cases, the switches must be unobstructed and clearly marked.
All Sun servers are shipped with a sufficient number of power supplies to provide all power needed by all Sun supported configurations.
Sun does not test many third-party products that are compatible with Sun servers. Therefore, Sun makes no representations about those products or about the power requirements for products not supplied by Sun.
Power constraints can occur in two areas:
To maintain a safe facility, you must ensure that the current draw does not exceed the maximum current limit for your power outlet. In the United States and Canada, the maximum is 80% of the outlet's total capacity. For areas outside of the United States and Canada, contact local agencies for information about local electrical codes.
See the site planning product specifications for the maximum input current and power consumption for your server.
Most servers shipped by Sun are configured with one or more power supplies, which are sufficient to support the maximum load of the server.
The servers provide "N+1" power supply redundancy to maintain system uptime. An N+1 redundant power supply configuration does not add to the power capacity of the servers. "N" represents the number of power supplies needed to power a fully configured server. The "1" means that there is one additional power supply in the server to handle the load if a supply fails. When the server is operating normally, all of the power supplies are turned on, even the redundant supplies.
In a 1+1 configuration (that is, two power supplies are installed, each capable of providing enough power for the entire server), both supplies are turned on and are delivering power. Each supply delivers approximately 50% of the power needed by the server. If one supply fails, the supply that is still online will deliver 100% of the power needed to keep the server running.
In a 2+1 configuration (that is, three power supplies are installed, with two power supplies delivering enough power for the entire server), all three power supplies are turned on and are delivering power. Each supply delivers approximately 33% of the power needed by the server. If one supply fails, the supplies that are still online will each provide 50% of the power needed to keep the server running.
Most power supplies cannot support the maximum values on all inputs at the same time because that would exceed the power supply's total output capacity. The load must be distributed among the power supplies in a way that does not exceed their individual maximum values.
The servers have built-in protection against exceeding the output capacity of the individual power supply. Be sure to consult the server documentation to learn how the servers will react during a power overload.
The PCI slots in the Sun servers comply with PCI Local Bus Specification Revision 2.1. The PCI buses in the servers are designed to provide either 15 watts or 25 watts of power, depending on the server, multiplied by the number of PCI slots in the PCI chassis. Thus, a 15-watt per slot, four-slot PCI chassis has a total of 60 watts of power available. These 60 watts can be used in any manner that conforms to the PCI standard. A single PCI slot can support a card that requires up to 25 watts. Each slot in the Sun Fire V490 server can supply up to 25 watts of power. The total power used by all six slots in the V490 must not exceed 90 watts.
Here are some examples of how you might populate a 60-watt, four-slot PCI chassis:
Servers and related equipment generate a considerable amount of heat in a relatively small area. This is because every watt of power used by a server is dissipated into the air as heat. The amount of heat output per server varies, depending on the server configuration.
The heat load in a data center is seldom distributed uniformly and the areas generating the most heat can change frequently. Further, data centers are full of equipment that is highly sensitive to temperature and humidity fluctuations.
See the site planning product specifications for your server's heat output, temperature, and humidity specifications.
Proper cooling and related ventilation of a server within a cabinet is affected by many variables, including the cabinet and door construction, cabinet size, and thermal dissipation of any other components within the cabinet. Therefore, it is the responsibility of the data center manager to ensure that the cabinet's ventilation system is sufficient for all the equipment mounted in the cabinet.
Do not use a server's nameplate power ratings when calculating the server's heat release. The purpose of the nameplate power ratings is solely to indicate the server's hardware limits for maximum power draw.
The flow of air through the server is essential to the proper cooling of the server. Even though the data center air might be at a safe and steady temperature at one location, the temperature of the air entering each server is critical. Problems sometimes arise for these reasons:
Most all Sun servers draw in ambient air for cooling from the front and discharge heated exhaust air to the rear. The servers require that the front and back cabinet doors to be at least 63% open for adequate airflow. This can be accomplished by removing the doors, or by ensuring that the doors have a perforated pattern that provides at least 63% open area. In addition, maintain a minimum of 1.5-inch (3.8-cm) clearance between the servers and front and back doors of a cabinet.
The servers are equipped with fans that route cool air throughout the chassis. As long as the necessary air conditioning is provided in the data center to dissipate the heat load, and sufficient space and door openings are provided at the front and back of the servers, the fans will enable the rackmounted servers to work within the operational temperature specifications. Again, see the site planning product specifications for your server's temperature specifications. See Cabinet Location for information about recommended placement of cabinets and racks to optimize proper aisle airflow.
A standard unit for measuring the heat generated within, or removed from, a data center is the British Thermal Unit (Btu). The heat produced by electronic devices such as servers is usually expressed as the number of Btu generated in an hour (Btu/hr).
Watts (W) is also a term used to express heat output and cooling. One watt is equal to 3.412 Btu/hr. For example, if you use 100 watts of power, you generate 341.2 Btu/hr.
Air conditioning capacity is also measured in Btu/hr or watts. Large air conditioning systems are rated in tons. One ton of air conditioning is a unit of cooling equal to 12,000 Btu/hr or 3517 watts.
The site planning product specifications provides the minimum, typical, and maximum heat output and cooling requirements for base configurations of your server. These specifications are the measured power ratings, which are calculated for the base server configurations as defined by Sun. It is important to realize the nameplate ratings are only a references to the servers' hardware limits that could accommodate future components. Do not use these values to calculate the servers' current power and cooling requirements.
In addition to the heat load generated by the servers, some cabinets include fans, power sequencers, and other devices that generate heat. Be sure to obtain the heat output values of these devices from your cabinet supplier. Also, when calculating data center cooling requirements, be sure to include heat dissipation for all equipment in the room.
To determine the heat output and cooling requirements of the rackmounted servers, add the Btu or watts for each server in the rack. For example, if one server is putting out 1000 Btu/hr (293 watts) and another one is putting out 2000 Btu/hr (586 watts), the total heat generated is 3000 Btu/hr (879 watts). The air conditioning equipment then should be properly sized to cool at least 3000 Btu/hr (879 watts) to accommodate these two servers.
See Calculating Cooling Requirements for an example of how to estimate cooling requirements based on the square footage used by the cabinets and racks in the data center.
In the book Enterprise Data Center Design and Methodology by Rob Snevely (available at http://www.sun.com/books/blueprints.series.html) the concept of using rack location units (RLUs) to determine heat output and cooling requirements in the data center is discussed. A rack location is the specific location on the data center floor where services that can accommodate power, cooling, physical space, network connectivity, functional capacity, and rack weight requirements are delivered. Services delivered to the rack location are specified in units of measure, such as watts or Btus, thus forming the term rack location unit.
Since today's data centers house hundreds or thousands of servers with widely varying power and cooling requirements, RLUs can help you determine where greater or less power and cooling services are needed. RLUs can also help you determine how to locate the racks to maximize services. Using square footage calculations for power and cooling assumes that power and cooling loads are the same across the entire room. Using RLUs lets you divide the data center into areas that need unique power and cooling services.
To determine RLUs for heat output and cooling, you must add together the heat output and cooling requirements for all servers installed in the rack. Then assess the RLUs for adjacent racks. For example, suppose you had 24,000 square feet of space in the data center. You might have a 12,000-square foot area where 600 PCs output 552,000 Btu/hour and need 46 Btu/hour of cooling per square foot. Another 6000-square foot area might contain 48 severs which output 1,320,000 Btu/hour and need 220 Btu/hour of cooling per square foot. A third 6000-square foot area might contain 12 high-end servers which output 972,000 Btu/hour and need 162 Btu/hour of cooling per square foot.
Using a square footage calculation for this example yields a cooling requirement for all three sections of 2,844,000 Btu/hour, or 118.5 Btu/hour of cooling per square foot. This would exceed the 46 Btu/hour cooling needed by the PCs, but it is much too little cooling capacity required for both server areas. Knowing the RLUs for power and cooling enable the data center manager to adjust the physical design, the power and cooling equipment, and rack configurations within the facility to meet the systems' requirements.