Design Tips for the Dual-Powered Data Center

1 comment

Brandon Perryman is Vice President of Sales and Marketing at Fibertown, a Tier IV-designed data center and business continuity campus. He leads the strategic direction for data center and business continuity office space sales and development into new geographic markets.

BRANDON PERRYMAN
Fibertown

Modern data centers using dual power designs have further increased reliability of IT systems by ensuring reliable power distribution and delivery. Simply buying dual-powered gear is not enough to ensure high availability and achieve the ultimate goal of zero downtime. The key is understanding power generation and delivery systems while avoiding four major design failures.

Understanding Power Generation and Delivery Systems

In a dual-powered data center, “concurrently maintainable” is achieved by delivering at least two power circuits (A and B) to each cabinet. If every piece of computer equipment is outfitted with dual power supplies, the load will continue to run even if one power source is shut down.

Failure to properly design, size and implement dual power infrastructure at the cabinet may lead to:

  • Breaker trip on failover
  • Breaker trip during restart
  • Power loss on single corded gear
  • Excessive power charges from under-utilization

1. Preventing Breaker Trip on Failover

In a dual system, power is typically supplied to the cabinet utilizing a whip. Whips are sized in increments of 10 amps. The standard allowable load on a whip is 80% of the breaker rating. For example, a 20 amp A-B whip pair would be limited to 16 amps.

Setting up A-B Power to Prevent Breaker Trip on Failover

Let’s say a server cabinet contains eight servers, each with dual power supplies consuming 2 amps per server at full running load. If a 120 volt, 20 amp A-B power whip pair is delivered to the cabinet the load will be distributed as follows:

  • Power Circuit A – 8 amps (with both A-B circuits active)
  • Power Circuit B – 8 amps (with both A-B circuits active)

Total power draw for the A and B circuits is 16 amps. If power circuit B fails or must be de-energized for maintenance, the A power circuit in all eight servers will be required to deliver twice the power to the server – for a total input load of 16 amps on a 20 amp power circuit.

Remember … 8 servers, each server needs 2 amps and only one power supply is now energized per server. This is within the 80% breaker rating design criteria, so this example is within the design specification.

When a breaker is tripped during failover, it’s usually caused by doubling the number of servers without accounting for power load. The 20 amp A circuit is loaded to 16 amps and the B circuit is also loaded to 16 amps.

If power circuit B fails or must be de-energized, the A power circuit in all 16 servers will be required to deliver twice the power. A total input load of 32 amps on a 20 amp power circuit. In this scenario, the breaker will trip and the servers will go down.

Preventing breaker trip during failover requires understanding your specific power load and accounting for the balance in your design.

2. Preventing Breaker Trip During Restart

Failure to properly design, size and implement dual power infrastructure at the cabinet may lead to breaker trip during restart. Improperly loaded circuits may support a running load in a failover situation, but the restarting of connected servers during single source operation could then trip the upstream circuit breaker.

Limiting current through the breaker to 80% of the breaker rating allows for surges of power to the load often experienced on start-up or other momentary loads. The 80% rating is a time versus temperature relationship, so the breaker is able to handle the start-up surge for a limited time, after which the loads return to normal.

In a failover situation, with one power circuit inoperative, the running load for one circuit in the example above would be 16 amps. If one of those devices were a large RAID array, the starting current could easily exceed 200% of the running load and exceed the breaker rating, thereby causing a breaker trip and resulting downtime.

3. Preventing Power Loss on Single Corded Gear

It’s good design principle to disallow the use of computer devices in a high-availability data center environment. However, some network products or legacy servers may only have single power supplies. These single power supply devices can still be used with reliability by utilizing automatic transfer switches.

These low-cost devices are typically rack mountable and occupy 1U or rack unit of space. They feature dual input cords and are able to switch from one power circuit to the other in a few micro seconds when power failure is detected on one of the input leads. This transfer time is typically well within the specification of most devices, so the blip is not seen by the load. The power fails, the load transfers and the attached devices continue operating normally.

4. Avoiding Excessive Power Charges from Under-utilization

Failing to fully load power circuits to their rated capacity may not result in downtime but could inflate power subscription costs. Avoid excessive power charges from under-utilization by loading each circuit to the rated capacity while respecting safety margins.
Another common pitfall is specifying the number of required circuits based on the number of PDUs or power trips in the cabinets. Data center power circuits are an expensive way to handle power distribution and care should be used to order only what is required.

While dual powered facilities provide concurrent maintainability and high availability when coupled with dual powered devices, proper power distribution planning is still required to achieve the ultimate goal of zero downtime and/or realize the cost benefits of outsourcing.

Industry Perspectives is a content channel at Data Center Knowledge highlighting thought leadership in the data center arena. See our guidelines and submission process for information on participating. View previously published Industry Perspectives in our Knowledge Library.

Add Your Comments

  • (will not be published)

One Comment

  1. Kyle Johnson

    The statement about rack mounted transfer switches, "They feature dual input cords and are able to switch from one power circuit to the other IN A FEW MICRO SECONDS when power failure is detected on one of the input leads" is not accurate. Mechanical contactor based switches can take up to 28 ms while solid state based transfer switches will be in the 4 ms range. There are also several other important considerations to be aware of when applying these devices.....