Data Management in the Cloud

Colloquium sur le Cloud Computing
Mardi 18 octobre 2011, 10h30-12h, Salle des Séminaires, LIRMM
Organisé par l’équipe Zenith (Esther Pacitti, Patrick Valduriez)

Amr El Abbadi

University of California, Santa Barbara

Abstract: Over the past two decades, database and systems researchers have made significant advances in the development of algorithms and techniques to provide data management solutions that carefully balance the three major requirements when dealing with critical data: high availability, reliability, and data consistency. However, over the past few years the data requirements, in terms of data availability and system scalability, from Internet scale enterprises that provide services and cater to millions of users has been unprecedented. Cloud computing has emerged as an extremely successful paradigm for deploying Internet and Web-based applications. Scalability, elasticity, pay-per-use pricing, and autonomic control of large-scale operations are the major reasons for the successful widespread adoption of cloud infrastructures. Current proposed solutions to scalable data management, driven primarily by prevalent application requirements, significantly downplay the data consistency requirements and instead focus on high scalability and resource elasticity to support data-rich applications for millions to tens of millions of users. However, the growing popularity of “cloud computing”, the resulting shift of a large number of Internet applications to the cloud, and the quest towards providing data management services in the cloud, has opened up the challenge for designing data management systems that provide consistency guarantees at a granularity which goes beyond single rows and keys. In this talk, we analyze the design choices that allowed modern scalable data management systems to achieve orders of magnitude higher levels of scalability compared to traditional databases. With this understanding, we highlight some design principles for data management systems that can be used to augment existing databases with new cloud features such as scalability, elasticity, and autonomy. We then present two systems that leverage these principles. The first system, G-Store, provides transactional guarantees on data granules formed on-demand while being efficient and scalable. The second system, ElasTraS, provides elastically scalable transaction processing using logically contained database partitions. Finally, we will present two techniques for on-demand live database migration, a primitive operation critical to provide lightweight elasticity as a first class notion in the next generation of database systems. The first technique, Albatross, supports live migration in a multitenant database serving OLTP style workloads where the persistent database image is stored in network attached storage. The second technique, Zephyr, efficiently migrates live databases in a shared nothing transactional database architecture.

Bio: Amr El Abbadi is currently Professor and Chair of the Computer Science Department at the University of California, Santa Barbara. He received his B. Eng. in Computer Science from Alexandria University, Egypt, and received his Ph.D. in Computer Science from Cornell University in August 1987. Prof. El Abbadi is an ACM Fellow. He has served as a journal editor for several database journals, including, currently, The VLDB Journal. He has been Program Chair for multiple database and distributed systems conferences, most recently SIGSPATIAL GIS 2010 and ACM Symposium on Cloud Computing (SoCC) 2011. He has also served as a board member of the VLDB Endowment from 2002—2008. In 2007, Prof. El Abbadi received the UCSB Senate Outstanding Mentorship Award for his excellence in mentoring graduate students. He has published over 250 articles in databases and distributed systems.

Distributed Web Search

Vincent Leroy
Yahoo Research
Barcelona, Spain

Abstract:
An appealing solution to scale Web search with the growth of the Internet is the use of distributed architectures. Distributed search engines rely on multiple sites deployed in distant regions across the world, where each site is specialized to serve queries issued by the users of its region.
Distributed search raises several challenges. In order to preserve the quality of the results, all documents should be taken into account during the evaluation of a query. However, for scalability reasons, each search site can only index a subset of the documents. When a user query requests a document which is not indexed locally, the search site has to contact the other sites in order to compute an exact result. As sites are distributed in different regions, this generates additional latency and reduces the satisfaction of the user.
In this presentation, I will describe the general architecture of a distributed search engine. Then, I will focus on the problem of assigning new documents to a search site

Salle des séminaires, LIRMM

Data Management for the Masses

Sihem Amer-Yahia
Yahoo Research
New York, USA

Abstract: The fast increasing content on collaborative tagging sites, such as Delicious, and collaborative rating sites, such as MovieLens, requires the development of scalable and efficient search and recommendation techniques. I will describe two concrete applications to illustrate the challenges behind data management for personalized content discovery on those sites. In the first application, network-aware search, I will argue that obvious adaptations of well-known top-k algorithms require to maintain per (seeker, keyword) indexes, due to the dependence of scores on the seeker’s interest network. I will therefore investigate two space-saving solutions for network-aware search, network clustering and behavior clustering. In the second application, group recommendation, the quality of a recommendation is a function of disagreement among group members. This calls for maintaining pair-wise user disagreement indexes. Therefore, I will explore two space-saving strategies for group recommendation, behavior factoring and partial materialization. I will show that scalable and efficient search and recommendation techniques rely on exploring the balance between storage volume and response time in both applications.

Acknowlegdements: Network-aware search is joint work with Michael Benedikt (Oxford), Laks Lakshmanan (UBC) and Julia Stoyanovich (UPenn). Group recommendation is joint work with Gautam Das (UT Arlington), Senjuti Basu Roy (UT Arlington) and Cong Yu (Google Inc)

Bio: Sihem has been a Senior Research Scientist at Yahoo! Labs since May 2006 after 7 years at AT&T Labs. She received her Ph.D. in CS from U. Paris-Orsay and INRIA, France. Sihem focuses on data management, query processing and relevance models that leverage social behavior for online content serving. Her professional activities include chairing the SIGMOD 2009 Tutorials, the Social Networks and Personal Information track at ICDE 2010, the Structured and Unstructured Data Track at WWW 2010, the SIGMOD 2011 Information Retrieval and Extraction track, and the EDBT/ICDT 2011 Industrial track. She is a member of the Board of Trustees of the VLDB Endowment and of the ACM SIGMOD Executive Committee. Sihem serves as the TODS, VLDB Journal, and Information Systems Journal Area Editor. She is currently visiting the Yahoo! Barcelona Lab.