Data Center Monitoring – How Much Is Enough?

You won't be surprised to learn that those of us at Packet Power are big advocates of data center power and temperature monitoring. After all, it is what we do.  

But how much monitoring do you need to do?  There are plenty of options, from monitoring every device to no monitoring whatsoever. We believe the right answer depends on what you want to get out of it -- and may involve fewer monitoring points than you might think.  

Start with what you want to do with the monitoring data.  Here are a few of the more typical options:

• Measure and track PUE.  Good news - this can be done with just two data points: total facility energy usage and IT equipment usage.  Mission accomplished with very little monitoring. But assuming you now want to improve your PUE, will you have gained enough insight to know how best to do it?  
• Prevent heat- and energy-related operational issues. Here monitoring can be useful in identifying potential problems such as hot spots and circuits that are approaching their utilization thresholds.  If your facility isn't changing too rapidly, you can deploy continuous monitoring on just the "likely suspects" - the most heavily loaded circuits, cabinets that produce a lot of heat, and difficult heat management areas such as cabinets at the ends of rows.  If you aren't already doing this type of monitoring, spending a few thousand dollars to keep a close eye on your trouble spots buys a lot of peace of mind.  
• Improve facility energy efficiency.  There are many ways to improve energy efficiency, and they have different implications for monitoring.  Looking to raise ambient air temperature? Comprehensive heat monitoring is important. Undertaking virtualization?  Monitor the gear that is involved - both what is being virtualized as well as the VM servers - at the cabinet level where possible and the device level where necessary.  Then save some money by redeploying the monitoring equipment to the gear involved in the next project.  Try to leave the circuit-level monitoring in place to make life easier down the road. 
• Allocate costs and carbon emissions.  Monitor at the circuit level where you can and at the device level only where usage at the circuit level needs to be subdivided. Leverage tools provided by the virtual machine platforms for allocating shared server usage. Take your PUE into account when doing your allocations. As more devices track their power usage directly and the software to gather and analyze that data continues to improve, managing this at the device level will become more feasible.  

Now for the easy question.  How often do you need to gather data?  Given the range of monitoring options available today, there is no reason to implement any type of monitoring that does not provide automated real-time collection and analysis of data. Thankfully, the days of taking meter readings by hand are rapidly fading.  

The bottom line: Knowing in advance how you will use energy usage information will help minimize your monitoring costs.  Get the ball rolling by picking a focus area, starting small and growing from there.