Justin J. Levandoski

LARS: A Location-Aware Recommender System

This paper proposes LARS, a location-aware recommender system that uses location-based ratings to produce recommendations. Traditional recommender systems do not consider spatial properties of users nor items, LARS, on the other hand, supports a taxonomy of three novel classes of location-based ratings, namely, spatial ratings for non-spatial items, non-spatial ratings for spatial items, and spatial ratings for spatial items. LARS exploits user rating locations through user partitioning, a technique that influences recommendations with ratings spatially close to querying users in a manner that maximizes system scalability while not sacrificing recommendation quality. LARS exploits item locations using travel penalty, a technique that favors recommendation candidates closer in travel distance to querying users in a way that avoids exhaustive access to all spatial items. LARS can apply these techniques separately, or in concert, depending on the type of location-based rating available. Experimental evidence using large-scale real-world data from both the Foursquare location-based social network and the Movie Lens movie recommendation system reveals that LARS is efficient, scalable, and capable of producing recommendations twice as accurate compared to existing recommendation approaches.

The Bw-Tree: A B-tree for new hardware platforms

The emergence of new hardware and platforms has led to reconsideration of how data management systems are designed. However, certain basic functions such as key indexed access to records remain essential. While we exploit the common architectural layering of prior systems, we make radically new design decisions about each layer. Our new form of B-tree, called the Bw-tree achieves its very high performance via a latch-free approach that effectively exploits the processor caches of modern multi-core chips. Our storage manager uses a unique form of log structuring that blurs the distinction between a page and a record store and works well with flash storage. This paper describes the architecture and algorithms for the Bw-tree, focusing on the main memory aspects. The paper includes results of our experiments that demonstrate that this fresh approach produces outstanding performance.

LARS*: An Efficient and Scalable Location-Aware Recommender System

This paper proposes LARS*, a location-aware recommender system that uses location-based ratings to produce recommendations. Traditional recommender systems do not consider spatial properties of users nor items; LARS*, on the other hand, supports a taxonomy of three novel classes of location-based ratings, namely, spatial ratings for non-spatial items, non-spatial ratings for spatial items, and spatial ratings for spatial items. LARS* exploits user rating locations through user partitioning, a technique that influences recommendations with ratings spatially close to querying users in a manner that maximizes system scalability while not sacrificing recommendation quality. LARS* exploits item locations using travel penalty, a technique that favors recommendation candidates closer in travel distance to querying users in a way that avoids exhaustive access to all spatial items. LARS* can apply these techniques separately, or together, depending on the type of location-based rating available. Experimental evidence using large-scale real-world data from both the Foursquare location-based social network and the MovieLens movie recommendation system reveals that LARS* is efficient, scalable, and capable of producing recommendations twice as accurate compared to existing recommendation approaches.

Identifying hot and cold data in main-memory databases

Main memories are becoming sufficiently large that most OLTP databases can be stored entirely in main memory, but this may not be the best solution. OLTP workloads typically exhibit skewed access patterns where some records are hot (frequently accessed) but many records are cold (infrequently or never accessed). It is more economical to store the coldest records on secondary storage such as flash. As a first step towards managing cold data in databases optimized for main memory we investigate how to efficiently identify hot and cold data. We propose to log record accesses - possibly only a sample to reduce overhead - and perform offline analysis to estimate record access frequencies. We present four estimation algorithms based on exponential smoothing and experimentally evaluate their efficiency and accuracy. We find that exponential smoothing produces very accurate estimates, leading to higher hit rates than the best caching techniques. Our most efficient algorithm is able to analyze a log of 1B accesses in sub-second time on a workstation-class machine.

RDF Data-Centric Storage

The vision of the Semantic Web has brought about new challenges at the intersection of web research and data management. One fundamental research issue at this intersection is the storage of the Resource Description Framework (RDF) data: the model at the core of the Semantic Web. We present a data-centric approach for storage of RDF in relational databases. The intuition behind our approach is that each RDF dataset requires a tailored table schema that achieves efficient query processing by (1) reducing the need for joins in the query plan and (2) keeping null storage below a given threshold. Using a basic structure derived from the RDF data, we propose a two-phase algorithm involving clustering and partitioning. The clustering phase aims to reduce the need for joins in a query. The partitioning phase aims to optimize storage of extra (i.e., null) data in the underlying relational database. Our approach does not assume a particular query workload, relevant for RDF knowledge bases with a large number of ad-hoc queries. Extensive experimental evidence using three publicly available real-world RDF data sets (i.e., DBLP, DBPedia, and Uniprot) shows that our schema creation technique provides superior query processing performance compared to state-of-the art storage approaches. Further, our approach is easily implemented, and complements existing RDF-specific databases.

Sindbad: a location-based social networking system

This demo presents Sindbad; a location-based social networking system. Sindbad supports three new services beyond traditional social networking services, namely, location-aware news feed, location-aware recommender, and location-aware ranking. These new services not only consider social relevance for its users, but they also consider spatial relevance. Since location-aware social networking systems have to deal with large number of users, large number of messages, and user mobility, efficiency and scalability are important issues. To this end, Sindbad encapsulates its three main services inside the query processing engine of PostgreSQL. Usage and internal functionality of Sindbad, implemented with PostgreSQL and Google Maps API, are demonstrated through user (i.e., web/phone) and system analyzer GUI interfaces, respectively.

FlexPref: A framework for extensible preference evaluation in database systems

Personalized database systems give users answers tailored to their personal preferences. While numerous preference evaluation methods for databases have been proposed (e.g., skyline, top-k, k-dominance, k-frequency), the implementation of these methods at the core of a database system is a double-edged sword. Core implementation provides efficient query processing for arbitrary database queries, however this approach is not practical as each existing (and future) preference method requires a custom query processor implementation. To solve this problem, this paper introduces FlexPref, a framework for extensible preference evaluation in database systems. FlexPref, implemented in the query processor, aims to support a wide-array of preference evaluation methods in a single extensible code base. Integration with FlexPref is simple, involving the registration of only three functions that capture the essence of the preference method. Once integrated, the preference method “lives” at the core of the database, enabling the efficient execution of preference queries involving common database operations. To demonstrate the extensibility of FlexPref, we provide case studies showing the implementation of three database operations (single table access, join, and sorted list access) and five state-of-the-art preference evaluation methods (top-k, skyline, k-dominance, top-k dominance, and k-frequency). We also experimentally study the strengths and weaknesses of an implementation of FlexPef in PostgreSQL over a range of single-table and multi-table preference queries.

Trekking through Siberia: managing cold data in a memory-optimized database

Main memories are becoming sufficiently large that most OLTP databases can be stored entirely in main memory, but this may not be the best solution. OLTP workloads typically exhibit skewed access patterns where some records are hot (frequently accessed) but many records are cold (infrequently or never accessed). It is still more economical to store the coldest records on secondary storage such as flash. This paper introduces Siberia, a framework for managing cold data in the Microsoft Hekaton main-memory database engine. We discuss how to migrate cold data to secondary storage while providing an interface to the user to manipulate both hot and cold data that hides the actual data location. We describe how queries of different isolation levels can read and modify data stored in both hot and cold stores without restriction while minimizing number of accesses to cold storage. We also show how records can be migrated between hot and cold stores while the DBMS is online and active. Experiments reveal that for cold data access rates appropriate for main-memory optimized databases, we incur an acceptable 7-14% throughput loss.

Skyline query processing for uncertain data

Recently, several research efforts have addressed answering skyline queries efficiently over large datasets. However, this research lacks methods to compute these queries over uncertain data, where uncertain values are represented as a range. In this paper, we define skyline queries over continuous uncertain data, and propose a novel, efficient framework to answer these queries. Query answers are probabilistic, where each object is associated with a probability value of being a query answer. Typically, users specify a probability threshold, that each returned object must exceed, and a tolerance value that defines the allowed error margin in probability calculation to reduce the computational overhead. Our framework employs an efficient two-phase query processing algorithm.

LLAMA: a cache/storage subsystem for modern hardware

LLAMA is a subsystem designed for new hardware environments that supports an API for page-oriented access methods, providing both cache and storage management. Caching (CL) and storage (SL) layers use a common mapping table that separates a pages logical and physical location. CL supports data updates and management updates (e.g., for index re-organization) via latch-free compare-and-swap atomic state changes on its mapping table. SL uses the same mapping table to cope with page location changes produced by log structuring on every page flush. To demonstrate LLAMAs suitability, we tailored our latch-free Bw-tree implementation to use LLAMA. The Bw-tree is a B-tree style index. Layered on LLAMA, it has higher performance and scalability using real workloads compared with BerkeleyDBs B-tree, which is known for good performance.