Database Seminar Series (2013-2014)

Unstructured text has long coexisted with relational data but has not been much treated as a first class citizen inside a relational database. Given that a large portion of text produced everyday in the public domain is primarily for human consumption, the data is often available in some form of natural language text. Searching such sources in granularities that are smaller than a document has been a challenge; yet querying for more factual information and relationships, or relating the encoded facts to more structured data provides many clear advantages.

In this talk, we give an overview of the work on natural language text as it pertains to querying. We showcase some of our work on the subject including some of the building blocks we have been working on querying, indexing and resolving queries. We conclude by presenting some of the challenges and open directions.

Umeshwar Dayal has recently joined Hitachi America Ltd., R&D Division, in Santa Clara, California, as Vice President, Big Data Lab, and Senior Fellow, Information Research. In this role, he is responsible for research and technology innovation in big data and advanced analytics. Prior to joining Hitachi America, Umesh was an HP Fellow and Director of the Information Analytics Lab at Hewlett-Packard Labs, Palo Alto, California, where he managed and led worldwide research programs focused on big data, real-time analytics, visualization, and business analytics solutions. Before joining HP, Umesh was a senior technical consultant at DEC's Cambridge Research Lab, Chief Scientist at Xerox Advanced Information Technology and Computer Corporation of America, and on the faculty at the University of Texas-Austin. Umesh received his PhD in Applied Mathematics from Harvard University, and his M.E. and B.E. degrees from the Indian Institute of Science, Bangalore, India.

Umesh is an ACM Fellow and a recipient of the Edgar F. Codd Award from ACM SIGMOD in 2010 for fundamental contributions to data management. He has chaired and served on the program committees of all the major database conferences. He has served on the Board of Trustees of the VDLB Endowment, and is currently an Editor-in-Chief of the VLDB Journal. He is a member of the Steering Committees of the IEEE International Conference on Data Engineering, the ER Conference on Conceptual Modeling, and the SPIE Conference on Visualization and Data Analysis. He is Chair of the Board of Directors of the NSERC Business Intelligence Network, a research network of Canadian universities and industry. He has published over 240 research papers and holds over 55 patents in the areas of data management, transaction processing, business process management, business intelligence, information visualization, and analytics solutions. He can be reached at Umeshwar.dayal@hal.hitachi.com.

Replicating data across multiple data centers does not only allow moving the data closer to the user and, thus, reducing latency for applications, but also increases the availability in the event of a data-center failure. It is therefore not surprising that companies like Google, Yahoo, and Netflix already replicate user data across geographically different regions. Replication across data centers, however, is expensive. Inter-data center network delays are in the hundreds of milliseconds and vary significantly. Synchronized wide-area replication is therefore often considered to be infeasible with strong consistency and current solutions either settle for asynchronous replication which implies the risk of losing data in the event of failures, restrict consistency to very small partitions or give up consistency entirely.

In this talk, I will first describe MDCC (Multi-Data Center Consistency), an optimistic 1-Phase Commit Protocol, that does not require a master or static partitioning, and can be used to achieve strong consistent replication at a cost similar to eventually consistent protocols. Afterwards, I will present our Predictive Latency-Aware NEtworked Transactions (PLANET), a new transaction programming model which empowers the application developer to cope with longer and unpredictable latencies caused by inter-data center communication.

Dr. Fox received his BSc in Computer Science from the University of Toronto in 1975 and his PhD in Computer Science from Carnegie Mellon University in 1983. In 1979 he was a founding member of the Robotics Institute of Carnegie Mellon University as well as the founding Director of the Intelligent Systems Laboratory within the Institute. He co-founded Carnegie Group Inc. in 1984, a software company that specialized in knowledge-based systems for solving engineering, manufacturing, and telecommunications problems, and was its Vice-President of Engineering and President/CEO. Carnegie Mellon University appointed him Associate Professor of Computer Science and Robotics in 1987 (with tenure in 1991). In 1988 he was the founding Director of the Center for Integrated Manufacturing Decision Systems at Carnegie Mellon. In 1991, Dr. Fox returned to the University of Toronto where he was appointed the NSERC Research Chairholder in Enterprise Integration and Professor of Industrial Engineering and Computer Science. In 1992, he was appointed Director of the Collaborative Program in Integrated Manufacturing. In 1993, Dr. Fox co-founded and was CEO Novator Systems Ltd., a pioneer in E-Retail software and services.

Dr. Fox's research led to the creation of the field of Constraint-Directed Scheduling within Artificial Intelligence, and several commercially successful scheduling systems and companies. He also pioneered the application of Artificial Intelligence to project management, simulation, and material design. He was the designer of one of the first commercial industrial applications of expert systems: PDS/GENAID, a steam turbine and generator diagnostic system for Westinghouse, which was a recipient of the IR100 in 1985 and is still in commercial use at Siemens. He was the co-creator of the Knowledge Representation SRL from which Knowledge Craft™ and ROCK™, commercial knowledge engineering tools, were derived, and KBS from which several commercial knowledge based simulation tools were derived. His current research focuses on the ontologies, common sense reasoning and their application to Smart Cities.

Dr. Fox was elected a Fellow of Association for the Advancement of Artificial Intelligence in 1991, a Joint Fellow of the Canadian Institute for Advanced Research and PRECARN in 1992, and a Fellow of the Engineering Institute of Canada in 2009. He is a past AAAI councillor, and a member of ACM and IEEE. Dr. Fox has published over 150 papers.

Paul (Per-Åke) Larson has conducted research in the database field for over 30 years. He was a Professor in the Department of Computer Science at the University of Waterloo for 15 years and joined Microsoft Research in 1996 where he is a Principal Researcher. Paul has worked in a variety of areas: file structures, materialized views, query processing, and query optimization among others. During the last few years he has collaborated closely with the SQL Server team on how to evolve the architecture of the core database engine for modern hardware.

In declarative data cleaning, data semantics are encoded as constraints and errors arise when the data violates the constraints. Various forms of statistical and logical inference can be used to reason about and repair inconsistencies (errors) in the data. Recently, unified approaches that repair both errors in data and errors in semantics (the constraints) have been proposed. However, both data-only approaches and unified approaches are by and large static in that they apply cleaning to a single snapshot of the data and constraints.

In this talk, I will present a continuous data cleaning framework that can be applied to dynamic data and constraint environments. Our approach permits both the data and its semantics to evolve and suggests repairs based on the accumulated evidence to date. Importantly, our approach uses not only the data and constraints as evidence, but also considers the past repairs chosen and applied by a user (user repair preferences).

I will then describe details of a repair classifier that predicts the type of repair needed to resolve an inconsistency and learns from past user repair preferences to recommend more accurate repairs in the future.

Power management, dark silicon, and partial failures mean that, in the future, computer hardware will most likely consist of a dynamically- changing set of heterogeneous processor cores. Contemporary operating system structures were not designed with this hardware model in mind, and have difficulty adapting to relatively simple concepts such as processor hotplug. Our work on meeting this challenge in the Barrelfish research OS has led us to treat cores as much as possible (but not entirely) like any other devices in the system. Several novel ideas make this possible: aside from the multikernel architecture itself, we leverage the externalization of kernel state through capabilities, and the concept of a "boot driver", which is the equivalent of a device driver for a processor core.

In this talk I will present our framework for managing a changing set of cores in a multikernel OS, and some of the surprising consequences: individual kernels can be rebooted, replaced, or upgraded on the fly, cores and hardware threads can be temporarily turned into coprocessors and back again, and per-core OS state can be quickly moved around the hardware to minimize energy usage or enforce performance guarantees.

Timothy Roscoe is a Professor in the Systems Group of the Computer Science Department at ETH Zurich. He received a PhD from the Computer Laboratory of the University of Cambridge, where he was a principal designer and builder of the Nemesis operating system, as well as working on the Wanda microkernel and Pandora multimedia system.  After three years building web-based collaboration systems at a startup company in North Carolina, Mothy joined Sprint's Advanced Technology Lab in Burlingame, California, working on application hosting platforms and networking monitoring.  Mothy joined Intel Research at Berkeley in April 2002 as a principal architect of PlanetLab, an open, shared platform for developing and deploying planetary-scale services. In September 2006 he spent four months as a visiting researcher in the Embedded and Real-Time Operating Systems group at National ICT Australia in Sydney, before joining ETH Zurich in January 2007.  He is a recipient of a 2013 ACM SIGCOMM 10-year test-of-time award, and a 2014 Usenix NSDI 10-year test-of-time award, both for his work on PlanetLab.  In 2014 he was made an ACM Fellow for his contributions to operating systems and networking research.  His current research interests include operating systems for heterogeneous multicore systems, and network architecture.

For the past few years, our group has been working on problems related to Big Data through several projects. After briefly discussing these projects, the rest of this talk will present DILoS, which focuses on load management for ``Big Streaming Data.''

Today, the ubiquity of sensing devices as well as of mobile and web applications continuously generates a huge amount of data in the form of streams, which need to be continuously processed and analyzed, to meet the near-real-time requirements of monitoring applications. Such processing happens inside Data stream management systems (DSMSs), which efficiently support continuous queries (CQs). CQs inherently have different levels of criticality and hence different levels of expected quality of service (QoS) and quality of data (QoD). In order to provide different quality guarantees, i.e., service level agreements (SLAs), to different client stream applications, we developed DILoS, a novel framework that exploits the synergy between scheduling and load shedding in DSMS. In overload situations, DILoS enforces worst-case response times for all CQs while providing prioritized QoD, i.e., minimize data loss for query classes according to their priorities. We further propose ALoMa, a new adaptive load manager scheme that enables the realization of the DILoS framework. ALoMa is a general, practical DSMS load shedder that outperforms the state-of-the-art in deciding when the DSMS is overloaded and how much load needs to be shed. We implemented DILoS in our real DSMS prototype system (AQSIOS) and evaluated its performance for a variety of real and synthetic workloads. Our experiments show that our framework (1) allows the scheduler and load shedder to consistently honor CQs' priorities and (2) maximizes the utilization of the system processing capacity to reduce load shedding.

Dr. Alexandros Labrinidis received his Ph.D degree in Computer Science from the University of Maryland, College Park in 2002. He is currently an associate professor at the Department of Computer Science of the University of Pittsburgh and co-director of the Advanced Data Management Technologies Laboratory (ADMT Lab). He is also an adjunct associate professor at Carnegie Mellon University (CS Dept).

Dr. Labrinidis' research focuses on user-centric data management for scalable network-centric applications, including web-databases, data stream management systems, sensor networks, and scientific data management (with an emphasis on big data). He has published over 70 papers at peer-reviewed journals, conferences, and workshops; he is the recipient of an NSF CAREER award in 2008. Dr. Labrinidis served as the Secretary/Treasurer for ACM SIGMOD and as the Editor of SIGMOD Record. He is currently on the editorial board of the Parallel and Distributed Databases Journal. He has also served on numerous program committees of international conferences/workshops; in 2014, he was the PC Track-Chair for Streams and Sensor Networks for the ICDE conference.

DILoS was developed in collaboration with Thao N.Pham (as part of her PhD thesis) and Panos K. Chrysanthis who is the director of the ADMT lab. This work has been funded in part by two NSF Awards and a gift from EMC/Greenplum.

Digital societies and markets increasingly mandate consistent procedures for the access, processing and storage of information. In the United States alone, over 10,000 such regulations can be found in financial, life sciences, health care and government sectors, including the Gramm-Leach-Bliley Act, the Health Insurance Portability and Accountability Act, the Sarbanes-Oxley Act, etc. A recurrent theme in these regulations is the need for regulatory compliant storage as an underpinning to ensure data confidentiality, access integrity and authentication; provide audit trails, guaranteed deletion, and data migration.

However, without the availability of practical, technology-backed enforcement solutions full regulatory compliance cannot be realized.

In this work we posit that the seemingly contradictory requirements of security, efficiency and low cost can in fact be reconciled gracefully via intelligent deployment of cryptographic and system security constructs.

To this end we design and prototype a number of fully functional relational database and file systems, addressing data privacy, query authentication and data retention, while offering increased functionality, higher efficiency and lower costs.

Radu is an Associate Professor of Computer Science at Stony Brook University (on leave) and currently the CEO of Private Machines Inc. He remembers when gophers were digging through the Internets and bits were running at slower paces of 512 per second. He is also interested in efficient computing with a touch of cyber-security paranoia, raising rabbits on space ships and sailing catamarans of the Hobie variety.