Compared to early Transaction Processing (TP) systems, today's TP systems have seen dramatic changes in both application requirements and underlying hardware support. As such, existing algorithms and design decisions hardly reflect reality and require revisiting them. However, more importantly, future generation TP systems must also necessarily evolve because of the emergence of energy efficiency as a first order design consideration.
This project is an early venture into a very new area, seeking to identify the most appropriate metrics for energy efficiency (QoE) in TP environments, the most appropriate ways to combine these energy efficiency metrics with existing quality of service (QoS) metrics and the most appropriate ways to specify any trade-off between QoE and QoS. Project plans also include the development of new scheduling algorithms and new TP system components that optimize a particular metric under different hardware configurations, as well as the experimental evaluation of the developed algorithms on simulation platforms and on real, state-of-the-art hardware.
This exploration potentially has great impacts in the development of new data management technologies. It is expected to advance the knowledge and understanding of the interplay among modern hardware components and facilitate the development of next generation TP systems that exploit new hardware features with the potential to achieve significant energy savings. This understanding could help formulate the foundations of the important area of energy-efficient data management, and thus, contribute to the societal goal of energy conservation and sustainability.
More information on the project can be found at www.energy-efficient-data-management.org/tps.
Energy consumption for computing devices in general and for data centers in particular is receiving increasingly high attention, both because of the increasing ubiquity of computing and also because of increasing energy prices and the pressing need for energy conservation. In this project, we investigated energy saving approaches in two data management environments – traditional data management and data stream management. In the context of traditional data management, our investigation in energy-efficient transaction processing led to the development of the QMD (Quasi Mirror Disks) scheme. QMD exploits flash technologies in existing storage systems (RAID) to improve energy efficiency in periods of low demand. Our evaluation using real workloads showed that QMD exhibits significant energy savings of up 31%. Theoretically, the energy savings in mirror-based RAID schemes can be at most 50%. Complex event detection over data streams has become ubiquitous through the widespread use of sensors, wireless connectivity and the wide variety of end-user mobile devices. Typically, such event detection is carried out by a data stream management system (DSMS) executing continuous queries (CQs), registered by the users. With the primary goal of reducing energy consumption at the mobile devices, we proposed BOSe*, a power-aware query operator placement algorithm that determines which part of a CQ execution plan should be executed at the DSMS and which part should be executed at the mobile devices. Our extensive experimental evaluation, using parameters obtained from experiments using real mobile devices, showed that BOSe* can achieve an overall energy reduction of up to 53%. As part of our investigation, we examined methods to economically harness off-grid renewable energy to support data centers and beyond. This resulted in a method to harness energy from isolated renewable energy resources by using the lifting power and energy-carrying capacity of hydrogen and by mapping this airship application to an existing heuristic approach in order to schedule each hydrogen-filled airship in a fleet to add optimum value or within a pre-determined factor of optimum value. We calculated that the excess lifting gas of our theoretical hydrogen airship and showed that it would surpass the FY2017 Department of Energy’s goal for hydrogen tube trailer delivery capacity by at least five times. Finally, this project offered research experience to two undergraduate students (REUs), led to three MS projects/degrees and engaged in research several PhD students either directly as research assistants or indirectly through PhD seminars. Two PhD seminars were developed and offered in the context of this award. In particular, the first one focused on Data Management on the Cloud and the second one on Scientific Databases.
