The information technology (IT) industry and companies like Amazon, Google and Microsoft have rapidly embraced a paradigm shift to cloud computing. Cloud environments provide the immense processing power needed to handle the vast amounts of transactions performed. However, maintaining the virtualized environment with its exponentially increasing demands means the pursuit of high-performance, low-cost technologies is ongoing.
Simon Fraser University (SFU) Computing Science Professor Jiangchuan Liu is at the forefront of new cloud service technologies. A Fellow of the Canadian Academy of Engineering and the Institute of Electrical and Electronics Engineers, Liu is an award-winning computer scientist who studies networking and multimedia systems with a particular interest in multimedia content processing and communications.
Chi Xu is a graduate of SFU and researches computer and network virtualization and performance issues in cloud computing. He also actively participates in open-source cloud application communities. Supervised by Professor Liu, he and fellow SFU researchers and doctoral students Xiaoqiang Ma and Haiyang Wang took a novel approach to enhancing the efficiency of the cloud. Their research collaboration also included SFU Computing Science Professor Ryan Shea. Their paper, Enhancing Performance and Energy Efficiency for Hybrid Workloads in Virtualized Cloud Environment, presents Hylics, a solution that enables quick and energy-saving execution of intense computational workloads.
We talked with Liu, Xu and Ma about their research.
Can you explain the rapid shift to cloud computing? Why is it a game-changer in the evolution of the IT industry?
Most of our work on cloud computing began before the pandemic, and we observed that many companies were reluctant to move their infrastructure to the cloud. However, the sudden and rapid shift to working from home with its increased technological demands meant that companies—including leaders like Amazon, Google and Microsoft—found the use of cloud computing essential.
The cloud helps speed up transaction times. For huge companies like Google and Microsoft with millions of customers to serve, the need for efficiency is key. With increased demand for these enterprise-level services—everything from core business functions such as e-commerce and social media, to data storage and maintenance, to user interactions, scaling services to handle peak loads, deep data analytics and security systems—companies had no choice but to transition to the cloud.
We are excited that our research has coincided with this paradigm shift and can help companies meet the technology demands presented by the trend of working from home.
What is a hybrid workload and how is it transforming IT?
As noted, cloud computing is a huge breakthrough in the evolution of the IT industry. With the paradigm shift to cloud computing, enterprise-level services are broken down into a collection of smaller independent workloads. All the workloads have their own logic and data, running on different software stacks—operating systems, middleware, runtime libraries and supporting applications. In this published research, these workloads are referred to as hybrid workloads, since they perform a mixture of computations, input-output (I/O) operations and network transmissions. Based on our observations, with the proper design, deploying hybrid workloads on the cloud has strong potential to improve the way it utilizes resources. This paves the way towards more energy efficient enterprise cloud operations, thereby reducing their climate footprint.
What is Hylics and how does this novel solution improve computational performance and efficiency?
Hylics is a virtualization software architecture that jointly optimizes I/O and computational performance for hybrid workloads. Diving deep into the virtualization subsystem design, our work is different from the previous research that focuses on I/O stack optimizations. The design goal of Hylics is to shorten the data traverse path for both computation and transmission. Meanwhile, Hylics also decouples computations and I/O operations, bringing less interference. The overall energy efficiency is subsequently improved. The Hylics design is not conﬁned to any speciﬁc platforms or workloads. Our long-term vision is to provide manageable, adaptable (or highly customizable) and environmentally sustainable computational and I/O services.
So not only does Hylics shorten processing times, it is also more energy efficient?
Energy-efficient enterprise cloud operations certainly reduce the carbon dioxide (CO2) emissions and climate footprint of IT services. In fact, the International Data Corporation shows that continued adoption of cloud computing could prevent the emission of more than one billion metric tons of CO2 over the next several years. This projection is based on the assumption that 60 percent of datacenters will adopt the technology and processes essential to become more sustainable and "smarter" by 2024.
We consider Hylics to be one such technology. In our experiments, Hylics reduced the power consumption of hybrid streaming workloads by up to 23 percent. This comparison is made when performing the identical workloads with or without Hylics. For hybrid file processing workloads, the power consumption is reduced by up to 44 percent.
Can you provide an update on your current work with Hylics?
We have implemented Hylics-based solutions on other popular virtualization platforms upon request, such as Xen and Docker platforms. Many interesting workloads, including deep learning, 3D reconstruction, and augmented and virtual reality are being actively incorporated to the Hylics optimization module, and have demonstrated considerable performance improvements in the production environments of the SFUCloud. Hylics has exhibited the potential to be an important part of the next-generation multimedia cloud architecture.
In future work, we plan to implement Hylics-based solutions on other virtualization platforms. Since Hylics significantly minimizes self-interference, we will also revisit the existing virtual machine resource allocation strategies to help cloud providers to achieve better service performance and cost efficiency.
To request a download of the source code please contact Professor Liu.
Remembering Ryan Shea
SFU Computing Science Professor Ryan Shea passed away in December 2019 after a remarkably brave battle with cancer. Shea is remembered as an integral part of the School of Computing Science, both as one of its top students and as an inspiring faculty member. He built SFUCloud from scratch, which became Computing Science’s experimental platform and is still used for research and teaching. He was also integral to the development of the Big Data Professional Master Program. His energy and enthusiasm for teaching, research and life is dearly missed.