This topic really interests me because it is the intersection of IT and power. IT completely depends on power, and these days we need to pay more attention to that. If we could make IT manage power conversely, that would be interesting and great. In Part 1 of this two-part blog, I will touch on power issues, including power pricing at data centers, to set the stage for software-defined power implementation, which I will discuss in Part 2. Can software-defined power solve data center power issues? Read on.
Power Shortage Problems
Power problems are increasing at data centers. A typical problem is insufficient power for operations. Your data center may experience a power shortage when:
- Your utility company does not have enough power to allocate.
- Your power system configuration is fixed and cannot be easily expanded.
There are two cases for #1. First, your area’s power consumption grows and the utility does not have the capacity to accommodate the growth. In this case, your utility company needs to improve transmission and distribution grids, including substations, to enhance the power capacity for your region. This is a very capital-intensive operation and may not take place soon, if ever. Second, your utility company has enough capacity most of the time but may run out of power when demand is high and everyone wants more power, as may happen during peak time.
In #2, the total amount of power you can draw is capped; you cannot draw more power beyond that. Because the power capacity and configuration for your data center is fixed, you need to make changes to increase power draw. The required changes are not confined to your data center. Your utility company may need to improve the distribution grid, including feeder lines along with equipment hanging and substations. In some cases, the transmission system may also need enhancements. These enhancements would take time, and the utility would charge you more for the new capacity.
In addition, your power system configuration is fixed not only for the data center as a whole but also for each section in it. If you add more IT equipment to a section that cannot accommodate the extra load, your PDU circuit breaker will trip. Even if you have enough rack space, you cannot place your new ICT gear because running it would cause a power shortage. This leaves you with stranded space. You have to rearrange and/or move your equipment, and that requires shutting down, powering down, unplugging, moving, replugging, powering up, and booting a lot of equipment. The process is tedious and the downtime for so much equipment impacts your business. Even before you do it, you need to review all the power requirements in your data center and make a plan to allocate ICT gear optimally. That is not an easy task.
For all the cases above, if higher power demand is permanent and constant, the time and cost may be well justified. But if the higher power draw is required only temporarily, such investments are not justified.
Power Charges for a Data Center
Nowadays, energy cost has become one of the major issues for data center operators. Even IT types can no longer ignore this issue. Power pricing for a data center is complex and varies from one region to another. The price has multiple attributes to consider, and every contract tends to be unique and therefore cannot be generalized. However, for the sake of this blog, I have simplified the story. Note that your mileage may vary according to your requirements, region, and contracts with your local utility company.
In most large commercial / industrial pricing structures , there are three main charges:
- connection – charged to use the distribution grid
- energy – charged for energy (kWh) consumption
- demand – charged for maximum power (kW) consumption
In general, if you require a large power draw, your utility company needs to prepare its distribution grid, including substations, to accommodate it. Obviously, if you require 50MW, the connection fee tends to be higher than when you require only 20MW, in general.
The energy charge is based on how much electric energy you consume. You’re used to that as a consumer: the more you use, the more you pay. In PG&E’s territory, large power consumers like data centers are on the time-of-use pricing structure. The per-kW charge varies according to the time of day and is typically highest during the afternoon (peak-time pricing). On top of that, a peak-day pricing surcharge is imposed on the days deemed to be peak days.
According to PG&E, demand charge is:
Many non-residential rates include a demand charge. Demand is a measurement of the highest usage of electricity in any single fifteen, (or sometimes five), minute period during a monthly billing cycle. Demand is measured in kilowatts (kW). High demand is usually associated with equipment start-up. By spreading equipment start-ups over a longer period of time, you may be able to lower demand and thereby reduce your demand charges.
If your load fluctuates and is spiky, you may have to pay a higher demand charge. Keeping your load balanced to avoid spikes is a good idea.
We can do a few things to minimize power charges at a data center:
- design and plan ahead to set the proper power capacity
- avoid a hefty demand charge by making loads flat
- be aware of peak-time and peak-day pricing
It is not easy to come up with the optimal power capacity. We tend to design for overcapacity, but if we plan to expand in the future, we may want to leave some headroom for growth. Many data centers may be classified as mission critical facilities, and controlling loads during high power-price time usually is not an option. Keeping loads flat is very hard to do, as we may not know the load trend ahead of time.
When to Control Power
Power, once generated, cannot be stored in mass and must be consumed right away. In other words, data center operators cannot control the power supply so we need to control the demand instead. The situation is the same at a data center. By control, I mean reduction in load demands. When is it time to reduce load demand? When the price is high and/or when there is not enough power. PG&E explains its demand-and response program as:
PG&E’s demand response programs are designed to enable customers to contribute to energy load reduction during times of peak demand. Most PG&E demand response programs also offer financial incentives for load reduction during times of peak demand.
Demand Response has been foreign to data centers because they are typically considered mission critical and cannot be shut down, no matter what. But loads at data centers primarily for labs and testing may be lowered as I’ve described here.
I will talk about one implementation of software-defined power in Part 2.
By the way, I would like to thank Peter Maltbaek of Power Assure for his insight into power-pricing issues at data centers.