The above graph shows power consumption (the upper line graph, denoted in Watts) plotted against occupancy levels. The premise behind U-CARE is that heating and cooling can be programmed according to predicted occupancy figures as modelled from past data. It is hoped that this model will result in energy savings. The data above clearly shows that there is no correlation between energy consumption and occupancy levels. A fundamental project expectation was that energy draw would be lower during higher periods of occupancy, since reduced heating levels would be required. This expectation has not been upheld by the data which indicates that heating/cooling is not triggered by ambient temperature. This proves great scope for improvement. So, will basing heating and cooling on predicted occupancy levels generate energy and cost savings?

One limitation of the metering data is that the BMS (Building Management System) does not facilitate the identification of when energy consumption is being used for heating and when it is being used for cooling. One of three scenarios is probable therefore – 1) the energy draw is primarily used for heating; 2) the energy draw is primarily used for cooling; 3) a mixture of heating/cooling is in operation, which could be rationalised as the two are essentially cancelling each other out within short time intervals.

This brings us to a third source of data from the U-CARE project – questionnaire responses. Users of Graham Hills 634 were invited to provide feedback on their comfort levels within the lab in relation to the existing heating/cooling settings, via an online questionnaire. Initial analysis indicates that the majority of users report that they are frequently too cold within the facility. This finding suggests that cooling is frequently operational and that it may account for the fairly steady power draw irrespective of the number of users in the room. Could it be therefore that current cooling levels could be cut? If it transpires that cooling is not in fact operational on a regular basis, accounting for the steady power draw, the conclusion will be that increased heating levels are required – this will definitely not result in energy savings!

Usage levels of all computer labs across Strathclyde’s campus is currently monitored; this data will give an indication of the number of people in a lab at any given time.

Assumptions

• 1 machine logged in = 1 person using machine
• Power down of monitor, power down of base unit = no one physically at machine

Some scenarios are not accounted for due to the above assumptions. For example, group work round a single machine, individuals in lab but not actively using a pc etc.

Data is being collected at one minute intervals. A .txt file is generated with the raw data,  giving details of every currently-logged-in computer. A summary .txt file is generated to provide aggregate data, for central labs only, containing lab name, number of PCs currently logged in, and total number of PCs.

U-CARE aims to establish a heating/cooling cycle based on predicted occupancy figures. We discovered that the control interface of the outdated air conditioning unit within our selected James Weir lab could not be directly interoperated with. As such, we decided to shift our site of investigation to a more modern facility within the Graham Hills building – lab 634 – in the hope of identifying opportunities for implementing energy savings. This involved re-installing the meter and spending time validating readings from it.

A fire in James Weir has since rendered our initially-selected lab unusable. This has in turn put pressure on alternative lab facilities across the campus so our occupancy data for Graham Hills 634 is likely to have become skewed due to irregular increased usage.

Inefficient cooling results in unnecessarily high levels of energy consumption. At Strathclyde the entire Graham Hills server room is continuously cooled; cool air is not restricted to cold aisles, nor is it directed to identified hot spots. Another problem at Strathclyde is that this server room is not nearly used to full capacity. These two factors would suggest that simple improvements could be made fairly easily. However, things are not that straight forward. The chillers cannot cope with any more hardware so we cannot simply increase the quantity of equipment being cooled within the current levels of air conditioning operating. We have to find other ways of creating savings.

Imperial College are looking at various cooling methods to try to achieve some savings. They are considering the following strategies, and a combination of them where feasible:
• Organise “cold aisles” and evacuation of hot air from “hot spots”
• Energy recovery system, free cooling
• Temperature adjustment of computer room AC units
• Adjustment of other parameters of AC units
• Smaller UPS unit
• Improved solar radiation control (windows films/coating)
• Improved air/coolant flow (filters, fans, pipes, ducts and ductwork)
• Improved lighting control (to be able to switch on and off selected areas only)
• Improved energy efficiency of lighting devices themselves
• Electrical mains transformer efficiency improvement
• Mains voltage adjustment (if feasible)

Others are investigating Liquid Cooling, which operates within a PC’s CPU as opposed to trying to cool it using an external source. “Just like a radiator for a car, a liquid cooling system circulates liquid through a heat sink attached to the processor inside of the computer. As the liquid passes through the heat sink, heat is transferred from the hot processor to the cooler liquid. This hot liquid then moves out to a radiator at the back of the case and transfers the heat to the ambient air outside of the case, rather than inside, causing significant damage. The cooled liquid then travels back through the system to the CPU to continue the process.” [Source: http://www.coolitsystems.com/index.php/en/about/featured-articles/118-why-liquid-cooling.html]

For the older estate, few of these are feasible in the short-term. Significant investment is required, for example, to set up a cold/hot aisle system or to install new easily controllable air conditioning units. We have discovered that most of these units (e.g. in the case of the Daiken, Toshiba and Hitachi models we have in place) it is not possible to interface directly with the control settings to implement, for example, customised heating/cooling programmes based on occupancy. With a fully operational facility already in place, purpose built a few years ago within a limited budget and with various practical constraints within the site, it is now difficult to find economically justifiable ways (both in terms of direct cost and staff implementation time) to make effective changes within existing resources.

That said, the calculated PUE (Power Usage Efficiency) and DCiE (Data Centre infrastructure Efficiency) indicate that the benchmark figures for the Graham Hills facility at Strathclyde is running close to efficient. These benchmarks measure efficiency in terms of IT equipment power consumption as compared with total facility power consumption. Although the relationship between these two measures indicates broad efficiency, it does not tell us whether or not the actual levels of energy used are low or high. Further research is needed to assess this.

There are some good news in this area. We started collecting automated readings for computer room electricity meters about a year ago and spent quite a lot of time getting all measurements collected, consolidated and reported consistently. There were some issues or faults with individual meters but now we have three months of good quality data for January, February and March.

The data matches manual readings and also gives readings at 30 min intervals throughout the day. Now when automated data can be trusted we can use the data for environment Dashboard to link energy consumption and service metrics. I started looking into service metrics that can represent demand for resources – number of servers, number of applications, number of systems etc. and ways of publishing a monthly report which we aim to distribute to ICT and systems owners. We do this for PCs to raise awareness and it makes sense for servers too.