2 core tenets of rack PDU redundancy

2 core tenets of rack PDU redundancy

Are you taking all the rights steps to keep the lights on in the data center?

Uninterruptible power supply (UPS) failures may be the top causes of data center downtime, but they're hardly the only causes. A failure at the power-distribution-unit (PDU) level can knock an entire row of cabinets offline, taking down mission-critical applications that disrupt business continuity. Either scenario boils down to a central precept of data center management: Operational integrity is born of redundancy.

Just as facility-wide outages can be salvaged by the UPS, power distribution faults should be preventable with the aid of PDU redundancy. Make sure that's the case by applying these two best practices for PDU redundancy. 

1. Two PDUs are always better than one

Starting at the most basic level, critical equipment should ideally be connected to two separate intelligent PDUs, with each being fed by different power sources. This ensures that electrical delivery doesn't miss a beat in the event that the primary power source goes offline. 

That said, two PDUs per cabinet in a row of eight cabinets necessitates running 16 total Ethernet cables to a network switch, which is a bit much. In addition to higher expenses for extra cabling, setup time is significant, which makes it difficult for facility staff to quickly add and remove equipment for the sake of IT flexibility. The alternative (not using PDUs with remote switching and power monitoring) isn't viable. Without power distribution data, you're basically operating on faith that power levels are safe and efficient. When as much as $8,851 per minute are at stake, that's not really an option. 

This is where daisy chaining comes into the picture. Best-in-class PDUs can be linked in succession via small cables, so that only the first PDU in a series is wired to the network switch. As a result, you can have redundant power distribution infrastructure without actually needing any additional networking cable. All PDUs are integrated through a single point of connection.

When PDUS fail, so does everything that is connected to them.When PDUs fail, so does everything that is connected to them.

2. RSTP is your VIP

"RSTP was designed to accommodate redundant links to networks."

The phrase "single point of connection" in the last paragraph may have given you pause, and for good reason. Relying on one point of connection for network integration is in direct violation of the notion that operational integrity is born of redundancy. If connectivity fails at any point in the chain, an entire row of PDUs can end up getting knocked offline. 

Enter Rapid Spanning Tree Protocol (RSTP). Unlike other connectivity protocols, RSTP was designed to accommodate redundant links to networks without creating switching loops, also known as broadcasting storms. In these loops, the message never really arrives, instead going in circles between the two connections and ultimately flooding the network. RSTP prevents that.

This is critical because it means a chain of PDUs using RSTP can be wired twice, once at the beginning of the series and again at the end. If a connection fails at any point of the chain, the other PDUs aren't affected since there is a second connection. This solves the problem of having to individually connect each PDU to the switch while providing redundancy – bringing you one step closer to true operational integrity.