Using the Internet of Things to manage your data centre

Understanding what sensors are available

The data centre is full of sensor technology. Servers, storage arrays, switches, routers, racks, aisles and data centre halls are bristling with technology. Historically these sensors are designed for different systems that are installed in the data centre and as such, are not part of a single solution.
The common sensor groups are:
 Servers: These are full of different
sensors that monitor the performance
of the hardware. They include the speed
of fans, input temperature, CPU
temperature and intrusion alarms.
 Storage systems: These contain the
same sensors that can be found in servers
- fan speed, input and CPU temperature,
intrusion alarms - but also have sensors
that monitor temperature across the drive
arrays.
 Switches and routers: Fan speed, input temperature, CPU temperature, number
and location of network ports in use, and
the throughput of network traffic.
 Power distribution units: Power
consumption, temperature, humidity/dew
point and airflow.
 Racks: These have become complex
systems with a very wide ranging set of
sensors. Input temperature at multiple
points such as bottom, middle and top,
airflow, intrusion alarms, smoke alarms,
CCTV, humidity/dew point and exhaust
temperature.
 Cooling systems: Temperature, airflow
and humidity/dew point.
 Aisles: Input temperature, airflow,
humidity/dew point, smoke/fire detection,
intrusion detection and CCTV
 Under floor: Water detection, airflow,
temperature and air pressure
 Above ceiling: Temperature, air pressure,
humidity/dew point
In addition to the physical sensors, there are a range of software tools that monitor usage. When their data is overlaid from that of physical sensors, it is possible to see how the impact of different workloads on the data centre can be modelled.

One monitoring solution to bind them all?
It’s a great idea and one that has been mooted many times but unfortunately there is no single solution that gathers all of the information from the wide array of sensors. There have been efforts to do this and a lot of progress has been made, especially in terms of monitoring the data centre infrastructure.
The one approach that has had good success is Data Centre Infrastructure Management (DCIM). According to analyst Gartner, there has been an explosion in the number of vendors offering DCIM solutions. By 2017, Gartner expects around 60% of data centres to be utilising some degree of DCIM solution. With over 75 vendors already offering their products as being DCIM compliant, it should raise alarms that it will take at least another two years to get to 60% coverage.
Some vendors offer a comprehensive management suite combined with physical sensors while others offer partial solutions often comprising a set of sensors. The bonus of being DCIM compliant is that a data centre operator can mix and match components into a larger system.

Other management solutions
The information from servers, storage systems and communication equipment such as switches and routers is often gathered into more traditional management suites such as HP OpenView and IBM Tivoli. These are not the only two suites, most hardware manufacturers have their own solutions for doing this.
These systems gather other information as well such as CPU, memory, disk and network usage. That information can be broken down by application and workload to show the impact of any particular workload across data centre or just one server or storage device.
Environmental information from outside of the data centre
Outside of the data centre there are other sensors that provide information that is rarely used. Airspeed indicators can be installed to show the direction and speed of the wind. This is important for those using free air cooling when they want to use the wind to help drive the fans that cool the data centre. It is also important for any facility that is using wind turbines to generate renewable energy systems that provide power to non critical systems such as lighting.
Wind speed and direction can also be combined with other factors such as pollen count. Free air cooling in areas close to high pollen crops such as oil seed rape has to be carefully monitored to prevent excess pollen and insect build up on the screens that protect the fans. It can also help when the data centre is located near to the coast as an onshore wind will increase the risk of salt water increasing the corrosion of equipment. This is information that can be used by maintenance teams to decide when systems will need preventative maintenance.
Temperature from outside weather stations is also important in free air cooling as it will determine the temperature differential between the exhaust and input air. Humidity sensors will also provide data on dew point and whether other systems will be required to remove/add humidity to the input air.

Analytics the key to gaining greater insight into the data centre
There are at least four distinct systems identified above - DCIM, traditional management tools, software management tools and external sensor data. Each of these systems can provide large amounts of data in real-time and the amount of data created in a mid-sized data centre, depending on the granularity of data, can be in the order of gigabytes of data per day.

There are various ways to use the data. The most efficient way is to us a transactional data base that captures the data using a common timeline as a reference code. That allows the data to be extracted and visualised to answer a number of key questions. Those questions may be operational or they may be financial.
One example of the operational data might be to look at what regular workloads are running over a set period of time. By identifying exactly how much resource such as CPU, disk and network are used, it is possible to model that workload. At this point the model can be used to see what would be the impact of changing the number of servers or places in the data centre where the workload is running.
There are several benefits that can be gain from this. It might be more cost effective to restrict where in the data centre the workload runs as this would reduce network traffic and the number of servers in use. The savings would be in the amount of power and cooling required for that workload.
No data centre runs a single workload therefore another benefit might be highlighting compatible workloads that can be run together on the same hardware rather than being spread across multiple systems.

While this would increase the usage of resources and the heat generated by those systems, it may have a low cost in terms of power and cooling than trying to spread the workload too widely across the data centre.
If a purely financial view such as lowering the cost of power and cooling is required, the analytics could be used to show how costs vary based on system loading across the data centre. This would highlight cost differences between different generations of hardware and even manufacturers. The result would be information that the buying department could use when negotiating the next round of hardware upgrades.
This same modelling approach can be combined with other data such as airflow through the data centre to see how a particular set of workloads and hardware generate heat. The data can be used as an input in to any data hall redesign or, when combined with external environmental information, where to consider locating a new data centre.

Why this creates an Internet of Things
What about that claim that the data centre is a good example of an Internet of Things (IoT)? A standard definition of IoT is multiple devices, connected together and sharing data without requiring human interaction.
DCIM, traditional management tools and software monitoring systems are all examples of IoT today. They can be scaled up to 10’s, 100’s or even 1000’s of devices. While there is some human interaction required initially to connect the systems together, once the system is created they can work autonomously.
Conclusion
The ability to take the three systems above and combine them into a new system that incorporates environment systems from outside of the data centre is an example of a new IoT solution. The environmental data can be used to decide when to activate or even deactivate free air cooling. Combining the data creates something that would not have existed before and can be used to better design data centres, manage cooling and even determine the location of data centres.
Despite all of the potential, less than 50 per cent of data centre owners and operators are currently using DCIM. Of these, only a very small percentage are combining it with other systems to help tune their power and cooling requirements. Despite the recent dip in oil prices, energy costs are predicted to continue to rise over the next decade so the sooner companies take advantage of all the systems and sensors at their command to help lower their costs, the better.