Real-time monitoring of power and energy will enable you to :
• Have the right power for the racks, depending upon the need : servers, workload. No lack, no excess.
• Optimize the rack density, and adapt it to the power available at the rack level. For example, place 20 servers, whereas you were initially thinking of a theorital capacity of 10 or 15. So you will optimize productivity. This will be true at rack level, row level, room level.
• Better balance power usage in the room, with single- or three-phase current, thanks to a much better distribution of the servers.
• Avoid power peaks
• Optimize cooling, with power usage and heat being homegeneous
• Take the most out of your existing infrastructure, and avoid over-investing
• Make sure that the contact with your power supplier really matches your need
So you will gain at twolevels : CAPEX, and OPEX.
Real-time monitoring of temperature will enable you to :
• Make sure that cooling matches the real need
• Quickly identify hot spots, if any
• Have a homogeneous temperature in the space being concerned, whatever the organization of your datacenter
Last, real-time monitoring of power and energy will enable you to :
• Anticipate on potential risks
• Plan and organize the evolution of your datacenter in the smartest way, starting from a complete and accurate knowledge of power & cooling needs.
NAMEPLATE POWER, DERATED POWER : STATIC POWER VERSUS REAL-TIME POWER !
A manufacturer generally gives a nameplate, static, power value to a server.
Most of the time, this value is a maximum one, not a realistic one.
So, if you have to cope with a static value, you will de-rate this value, and consider then a static, de-rated value.
Even if closer to reality, the de-rated value does not match reality.
Let us take one example :
• A server has a 500 W nameplate power
• You decide, because you have no real-time monitoring, yet, that you derated this value, with a 80% factor : you then decide to consider a 400 W derated value for all your calculations & estimations
• In fact, a real-time monitoring tells that the same server needs, on an average basis, 300 W
• So, if you have 6000 X for the rack, depending upon your way to meter, the number of servers being powered and placed may change significanttly :
- nameplate power : 12 servers possible
- derated power : 15 servers possible
- real-time power, with average being 300 W : 20 servers possible.
• Therefore, depending upon the way power is estimated or measured, you have the capacity for 5 more servers (real-time compared to de-rated) or 8 new servers (real-time compared to name-plate)
• Your gain is two-fold : CAPEX (higher ROI on rack & infrastructure) & OPEX (more workload for the same power).