When trying to describe the rapid pace of digital data growth – the cloud is said to be doubling roughly every two years – observers invariably reach for words like “mind-boggling,” “exponential” and “explosive.” Understandably, given the volumes, storage is a constant preoccupation. Yet there’s actually another pressing challenge associated with this data tsunami that isn’t talked about quite as much – namely, energy consumption. Fact is though, all that data – all our documents and Instagram posts, reams of instant messaging, audio and video streaming and more – is having an increasingly negative environmental impact.
According to one 2015 study, in the last century the expansion in IT technology caused energy consumption to surge, a trend that continued into this century. In 2012 for example, information and communications technology (ICT) gobbled up roughly 4.7 per cent of the world’s electrical energy, and accounted for around 1.7 per cent of total CO2 emissions.
In Ryerson’s Computer Science department, Zainab Al-Zanbouri, who’s working towards a PhD at the moment, is passionate about green IT solutions. Recently, supervised by Professor Cherie Ding, Al-Zanbouri turned her attention to data mining. “Establishing a quality of service (QoS)-based web service selection approach that considers energy consumption as one of the essential QoS properties represents a significant step towards selecting the greener web service,” she says, noting that currently most QoS-based web service selection approaches don’t consider energy consumption levels at all. “There is a need to identify services that consume less energy. If we can find out the general patterns of the relationship between energy consumption and dataset attributes, we can make a positive contribution to the research on selecting green services.”
Zainab presenting her paper at SCC in June, 2017
Describing her research in more detail, Al-Zanbouri says, “Our study shows that there is a strong relation between the dataset properties such as dataset size, number of attributes, data type, and QoS attributes energy consumption and latency.” For example, she says, “A service may consume more energy when processing a big dataset and less energy on a smaller dataset, or more energy on nominal dataset and less energy on numeric dataset.”
Al-Zanbouri has built a prediction system and used it to predict the energy consumption and latency values for data mining web services on a given dataset. “In this work,” she says, “we have tested four different prediction algorithms – linear regression, M5P, multi-layer perceptron and REPTree from Weka... Then using the predicted QoS values, we can rank the services. The ranking list can help users select services with good conventional QoS values (i.e. latency) and good energy consumption values.”
Ultimately, Al-Zanbouri says her experimental results showed the effectiveness of the prediction and service selection system. She recently had the honour of presenting her research in Honolulu, Hawaii at the 2017 IEEE International Conference on Services Computing in a paper called, “Selecting Green Data Mining Services.”
Al-Zanbouri, who previously earned her MSc in Computer Science at Ryerson, values the knowledge and experience she’s gained during her graduate studies. “One of the things I really appreciate is working with supervisors who have a lot of experience in the corresponding research area which really helped me to understand and do my work successfully. Moreover, I appreciate all the support we get from our professors and the staff in the Computer Science department.”
Al-Zanbouri’s research received funding from NSERC, Ryerson’s Computer Science department, the YSGS Grad Student Fund and the RSU Graduate Council.