We live in a data-driven world as everyone around has been telling us for some time. Everything is generating data, in volumes and at high rates, from the sensors embedded in our physical spaces to the large number of machines in data centers which are being monitored for a wide variety of metrics. The question that we pose is: Can all this data be used for improving the dependability of computing systems? Dependability is the property that a computing system continues to provide its functionality despite the introduction of faults, either accidental faults (design defects, environmental effects, etc.) or maliciously introduced faults (security attacks, external or internal). We have been addressing the dependability challenge through large-scale data analytics applied end-to-end from the small (networked embedded systems, mobile and wearable devices) [UsenixSec-20, DSN-19, UsenixSec-18, DSN-18, S&P-17] to the large (edge and cloud systems, distributed machine learning clusters) [ICS-19, ICST-19, TDSC-18, CGO-17, DSN-17]. In this talk, I will first give a high-level view of how data analytics has been brought to bear on dependability challenges, and key insights arising from our work. Then I will do a deep dive into the problem of configuring complex systems to meet dependability and performance requirements, using data-driven decisions.
Reconfiguring distributed NoSQL databases under changing workload patterns is crucial for maximizing database throughput. This is challenging because of the large configuration parameter search space with complex interdependencies among the parameters. While state-of-the-art systems can automatically identify close-to-optimal configurations for static workloads, they suffer for dynamic workloads as they overlook the performance degradation during the reconfiguration process and often violate the application’s availability requirements during reconfiguration. I will present our solution Sophia, which addresses these problems, with instantiation for NoSQL databases, Cassandra and Redis [UsenixATC-19, Middleware-17, BriefingsinBioinformatics-17].
Videos take lot of time to transport over the network, hence running analytics on live video at the edge devices, right where it was captured has become an important system driver. However these devices, e.g., IoT devices, surveillance cameras, AR/VR gadgets are resource constrained. This makes it impossible to run state-of-the-art heavy Deep Neural Networks (DNNs) on them and yet provide low and stable latency under various circumstances, such as, changes in the resource availability on the device, the content characteristics, or requirements from the user. I will introduce ApproxNet, a video analytics system for the edge, which makes content-dependent approximations, in a manner that achieves the desired inference latency versus accuracy trade-off under different system conditions and resource contentions, variations in the complexity of the video contents and user requirements [arXiv-19, UsenixATC-18, Middleware-18].
Saurabh Bagchi is a Professor in the School of Electrical and Computer Engineering and the Department of Computer Science at Purdue University in West Lafayette, Indiana. He is the founding Director of a university-wide resiliency center at Purdue called CRISP (2017-present) and co-lead on the WHIN center for IoT testbeds for digital agriculture and advanced manufacturing. He is the recipient of the Alexander von Humboldt Research Award (2018), an Adobe Research award (2017), the AT&T Labs VURI Award (2016), the Google Faculty Award (2015), and the IBM Faculty Award (2014). He serves on the IEEE Computer Society Board of Governors. Saurabh's research interest is in distributed systems and dependable computing. He is proudest of the 21 PhD and about 50 Masters students who have graduated from his research group and who are in various stages of building wonderful careers in industry or academia. In his group, he and his students have far too much fun building and breaking real systems for the greater good. Saurabh received his MS and PhD degrees from the University of Illinois at Urbana-Champaign and his BS degree from the Indian Institute of Technology Kharagpur, all in Computer Science.
Faculty Host: Matthew Caesar