According to Moore's Law, computing capacities will double every 18 months. If we apply this law to the average data center, it means that a data center's power requirements also tend to increase every year (if not double as such). Now the question arises as to how IT teams should size their organizations' power infrastructure.
The first step to sizing power infrastructure is to calculate the data center power requirements. There are different methods to determine your data center power requirements, depending on whether it is a professional internet data center or a captive data center. According to C. Kajwadkar, the chief architect and vice president of Netmagic Solutions, the first step is to consider the present and future power requirements of the servers. "The IT team should estimate the entire ICT infrastructure's present power requirements, keeping in mind the factor of future expandability. Also take into consideration the illumination and PC loads," Kajwadkar says. Calculation of server power consumption, the entire IT infrastructure's aggregate power requirements, and future power requirements can be obtained from hardware vendors' specifications. Abhay Johorey, chief operating officer of Aviva Life Insurance India, uses a methodology that takes a uniform load distribution across racks and cubic feet space available in data center for cooling. The load is calculated at the actual rated capacity for other equipment.
G.B. Ravichandra, the country sales director for enterprise business of APC India, points out that its essential to understand limitations (current and impending) during power sizing. "With the dearth of space and power, IT teams should look at how much of these commodities will be available. Accordingly, each team can chalk out the data center's power strategies," says Ravichandra.
The UPS angle in power sizing
Even if calculation of various power aspects is conducted to the minutest detail, everything goes for a toss if one does not size the uninterruptible power supply (UPS) properly or lacks a proper backup. Traditionally, the total number of servers that run on the UPS is taken into account for UPS sizing. IT infrastructure will always be a victim of changing loads, and this eventually depends on the UPS for power supply. The data center's cooling loads such as air conditioning (AC) also contributes towards a significant power load. "There is a design parameter that the UPS and AC should be sized to a certain extent. You cannot cross the installed capacity of the UPS and ACs at any given point of time. There are technologies that will allow you to give a flexible output of both UPS and AC within that limit. A few UPSs should be running in parallel and a few turned off. Same will be the case with ACs," says Kajwadkar.
To make sure that the data center does not come to a halt, UPS redundancy should be in place. There are various possible topologies and configurations to ensure data center uptime. "For critical data centers, an N+N configuration is recommended. It is a dual base architecture. It means two UPS systems are working independently and there is a fault tolerance design, which means anything happening to one UPS will not affect the other UPS systems," says Venkat Rao, the country manager for medium and large UPS systems at Emerson Network Power (India) Pvt Ltd.
In the case of Aviva, Johorey and his team follow an n+1 redundancy model. In this architecture, two redundant UPSs feed two different power distribution units on data center racks. Each UPS consists of a group of two UPS units, which ensures the second level of redundancy.
Cooling and data center power
There's a popular school of thought, which believes that proper UPS sizing and redundancy should be coupled with proper cooling and airflow to reduce data center power consumption. However, there are opposing views on this front.
According to Rao, smooth airflow offers nothing to aid power consumption. "Smooth airflow fits more in line with the data center's overall requirements. Servers perform many computing activities, for which it has to consume a certain amount of power. Air movement ensures that the servers are kept at a particular temperature," says Rao.
Giving us a different view, Ravichandra says that servers do not consume power in a structured fashion. "Servers with higher loads require a higher power feed, as compared to other servers. These will require dedicated in-rack cooling, since it is not advisable that the entire data center has to be cooled to cater to these servers," says Ravichandra.
Efficient power management is the last lap to cover on the power infrastructure sizing front. When it comes to this aspect, there's significant scope for innovation and experiments. For example, Netmagic moved from CFL to LED lighting for illumination. Also, Netmagic uses copper cables instead of aluminum. Reduction of heat, lesser power conditioning equipment losses, and efficient distribution inside the data center also contribute to lesser power consumption at Netmagic. "We have several racks, and feed right sized power to each rack. We manage power at the rack level rather than at a data center level. A power-efficient UPS for superior power usage effectiveness is also on the plans. Our AC will also undergo design changes to reduce our overall requirement of power capacity for data center cooling. " says Kajwadkar.
Aviva uses in-row cooling solutions for rack level cooling. "We do not waste power for cooling non-utilized space or racks. Also, we use virtualization to accrue server power utilization savings," says Johorey.