The rackable wiring quandary

Many data centers were designed to provide floor-based plenum cooling in an era where a 42U rack might have to dissipate just 3KW of server power. In such a rack, typically a maximum of 42U can be used, and a rack with 40 servers (with 2U reserved for Ethernet switches) means that the power requirement must be made for 20KW -- nearly seven times the load the data center may have been designed to both provide power for, and to cool.

In addition, 40U of rackable servers can mean a minimum of 40 power cables (double that number for redundant power sourcing, for a total of 80 power cables) and at minimum, 40 Ethernet cables (and the two-24-port or one-48 port Gigabit Ethernet switch needed to link them in a network). This produces a spaghetti-of-cables effect that also can block some of the airflow needed to chill the cabinet adequately. Headless operation is typical; that is no monitor/keyboard/mouse. But accessibility for each server becomes difficult in terms of requiring additional equipment (a ladder, as an example) to service a single rackable server.

The same design characteristic of older data centers also puts a certain ceiling on the number of blade servers that can be installed into a single rack, unless power and cooling components can be retrofit. IBM and third-party vendors, such as APC, Emerson and Leibert, offer substantial options to retrofit data centers to manage very high power loads at an additional cost.

To continue reading, register here and become an Insider. You'll get free access to premium content from CIO, Computerworld, CSO, InfoWorld, and Network World. See more Insider content or sign in.

Many data centers were designed to provide floor-based plenum cooling in an era where a 42U rack might have to dissipate just 3KW of server power. In such a rack, typically a maximum of 42U can be used, and a rack with 40 servers (with 2U reserved for Ethernet switches) means that the power requirement must be made for 20KW -- nearly seven times the load the data center may have been designed to both provide power for, and to cool.

In addition, 40U of rackable servers can mean a minimum of 40 power cables (double that number for redundant power sourcing, for a total of 80 power cables) and at minimum, 40 Ethernet cables (and the two-24-port or one-48 port Gigabit Ethernet switch needed to link them in a network). This produces a spaghetti-of-cables effect that also can block some of the airflow needed to chill the cabinet adequately. Headless operation is typical; that is no monitor/keyboard/mouse. But accessibility for each server becomes difficult in terms of requiring additional equipment (a ladder, as an example) to service a single rackable server.

The same design characteristic of older data centers also puts a certain ceiling on the number of blade servers that can be installed into a single rack, unless power and cooling components can be retrofit. IBM and third-party vendors, such as APC, Emerson and Leibert, offer substantial options to retrofit data centers to manage very high power loads at an additional cost.

Some vendors have addressed the problem by offering oversized racks with 45, 48 and custom-unit-size displacements. Rack depth is increasing to accommodate deeper chassis, both for blade servers and for rack-mounted systems.

Highly articulate cooling schemes also have become available to aid front-to-back cooling, rather than use traditional floor-based chilled air plenums -- so that heat dispersion can be focused for high cubic-feet-per-minute chilling loads needed by blades, as well as other high-density chassis supporting routers, switches and VoIP gateway telephony components.

< Return to main test

Read more about data center in Network World's Data Center section.