In this paper, we look back at the evolution of Yahoo!'s geo-replicated cloud data store from a research project called PNUTS to a globally deployed production system called Sherpa, share some of the lessons learned along the way, and finally, compare PNUTS with current operational cloud stores.

We consider the problem of resolving duplicates in a database of places, where a place is defined as any entity that has a name and a physical location. When other auxiliary attributes like phone and full address are not available, deduplication based solely on names and approximate location becomes an exceptionally challenging problem that requires both domain knowledge as well an local geographical knowledge. For example, the pairs "Newpark Mall Gap Outlet" and "Newpark Mall Sears Outlet" have a high string similarity, but determining that they are different requires the domain knowledge that they represent two different store names in the same mall. Similarly, in most parts of the world, a local business called "Central Park Cafe" might simply be referred to by "Central Park", except in New York, where the keyword "Cafe" in the name becomes important to differentiate it from the famous park in the city.

De-duplication - identification of distinct records referring to the same real-world entity - is a well-known challenge in data integration. Since very large datasets prohibit the comparison of every pair of records, blocking has been identified as a technique of dividing the dataset for pairwise comparisons, thereby trading off recall of identified duplicates for efficiency. Traditional de-duplication tasks, while challenging, typically involved a fixed schema such as Census data or medical records. However, with the presence of large, diverse sets of structured data on the web and the need to organize it effectively on content portals, de-duplication systems need to scale in a new dimension to handle a large number of schemas, tasks and data sets, while handling ever larger problem sizes. In addition, when working in a map-reduce framework it is important that canopy formation be implemented as a hash function, making the canopy design problem more challenging. We present CBLOCK, a system that addresses these challenges.

In this demonstration, we showcase the technologies that we are building at Yahoo! for Web-scale Information Extraction. Given any new Website, containing semi-structured information about a pre-specified set of schemas, we show how to populate objects in the corresponding schema by automatically extracting information from the Website.

Often an interesting true value such as a stock price, sports score, or current temperature is only available via the observations of noisy and potentially conflicting sources. Several techniques have been proposed to reconcile these conflicts by computing a weighted consensus based on source reliabilities, but these techniques focus on static values. When the real-world entity evolves over time, the noisy sources can delay, or even miss, reporting some of the real-world updates. This temporal aspect introduces two key challenges for consensus-based approaches: (i) due to delays, the mapping between a source's noisy observation and the real-world update it observes is unknown, and (ii) missed updates may translate to missing values for the consensus problem, even if the mapping is known. To overcome these challenges, we propose a formal approach that models the history of updates of the real-world entity as a hidden semi-Markovian process (HSMM). The noisy sources are modeled as observations of the hidden state, but the mapping between a hidden state (i.e. real-world update) and the observation (i.e. source value) is unknown. We propose algorithms based on Gibbs Sampling and EM to jointly infer both the history of real-world updates as well as the unknown mapping between them and the source values. We demonstrate using experiments on real-world datasets how our history-based techniques improve upon history-agnostic consensus-based approaches.

Computing interesting measures for data cubes and subsequent mining of interesting cube groups over massive data sets are critical for many important analyses done in the real world. Previous studies have focused on algebraic measures such as SUM that are amenable to parallel computation and can easily benefit from the recent advancement of parallel computing infrastructure such as MapReduce. Dealing with holistic measures such as TOP-K, however, is nontrivial. In this paper, we detail real-world challenges in cube materialization and mining tasks on web-scale data sets. Specifically, we identify an important subset of holistic measures and introduce MR-Cube, a MapReduce-based framework for efficient cube computation and identification of interesting cube groups on holistic measures. We provide extensive experimental analyses over both real and synthetic data. We demonstrate that, unlike existing techniques which cannot scale to the 100 million tuple mark for our data sets, MR-Cube successfully and efficiently computes cubes with holistic measures over billion-tuple data sets.

Complex information extraction (IE) pipelines are becoming an integral component of most text processing frameworks. We introduce a first system to help IE users analyze extraction pipeline semantics and operator transformations interactively while debugging. This allows the effort to be proportional to the need, and to focus on the portions of the pipeline under the greatest suspicion. We present a generic debugger for running post-execution analysis of any IE pipeline consisting of arbitrary types of operators. For this, we propose an effective provenance model for IE pipelines which captures a variety of operator types, ranging from those for which full to no specifications are available. We have evaluated our proposed algorithms and provenance model on large-scale real-world extraction pipelines.

Cube computation over massive datasets is critical for many important analyses done in the real world. Unlike commonly studied algebraic measures such as SUM that are amenable to parallel computation, efficient cube computation of holistic measures such as TOP-K is non-trivial and often impossible with current methods. In this paper we detail real-world challenges in cube materialization tasks on Web-scale datasets. Specifically, we identify an important subset of holistic measures and introduce MR-Cube, a MapReduce based framework for efficient cube computation on these measures. We provide extensive experimental analyses over both real and synthetic data. We demonstrate that, unlike existing techniques which cannot scale to the 100 million tuple mark for our datasets, MR-Cube successfully and efficiently computes cubes with holistic measures over billion-tuple datasets.

Automatic extraction of structured records from inconsistently formatted lists on the web is challenging: different lists present disparate sets of attributes with variations in the ordering of attributes; many lists contain additional attributes and noise that can confuse the extraction process; and formatting within a list may be inconsistent due to missing attributes or manual formatting on some sites.

Knowledge bases of entities and relations (either constructed manually or automatically) are behind many real world search engines, including those at Yahoo!, Microsoft, and Google. Those knowledge bases can be viewed as graphs with nodes representing entities and edges representing (primary) relationships, and various studies have been conducted on how to leverage them to answer entity seeking queries. Meanwhile, in a complementary direction, analyses over the query logs have enabled researchers to identify entity pairs that are statistically correlated. Such entity relationships are then presented to search users through the "related searches" feature in modern search engines. However, entity relationships thus discovered can often be "puzzling" to the users because why the entities are connected is often indescribable. In this paper, we propose a novel problem calledentity relationship explanation, which seeks to explain why a pair of entities are connected, and solve this challenging problem by integrating the above two complementary approaches, i.e., we leverage the knowledge base to "explain" the connections discovered between entity pairs. More specifically, we presentREX, a system that takes a pair of entities in a given knowledge base as input and efficiently identifies a ranked list of relationship explanations. We formally define relationship explanations and analyze their desirable properties. Furthermore, we design and implement algorithms to efficiently enumerate and rank all relationship explanations based on multiple measures of "interestingness." We perform extensive experiments over real web-scale data gathered from DBpedia and a commercial search engine, demonstrating the efficiency and scalability ofREX. We also perform user studies to corroborate the effectiveness of explanations generated byREX.

We make the case for developing a web of concepts by starting with the current view of web (comprised of hyperlinked pages, or documents, each seen as a bag of words), extracting concept-centric metadata, and stitching it together to create a semantically rich aggregate view of all the information available on the web for each concept instance. The goal of building and maintaining such a web of concepts presents many challenges, but also offers the promise of enabling many powerful applications, including novel search and information discovery paradigms. We present the goal, motivate it with example usage scenarios and some analysis of Yahoo! logs, and discuss the challenges in building and leveraging such a web of concepts. We place this ambitious research agenda in the context of the state of the art in the literature, and describe various ongoing efforts at Yahoo! Research that are related.

On script-generated web sites, many documents share common HTML tree structure, allowing wrappers to effectively extract information of interest. Of course, the scripts and thus the tree structure evolve over time, causing wrappers to break repeatedly, and resulting in a high cost of maintaining wrappers. In this paper, we explore a novel approach: we use temporal snapshots of web pages to develop a tree-edit model of HTML, and use this model to improve wrapper construction. We view the changes to the tree structure as suppositions of a series of edit operations: deleting nodes, inserting nodes and substituting labels of nodes. The tree structures evolve by choosing these edit operations stochastically.

The rapid growth of Web communities has motivated many solutions for building community data portals. These solutions follow roughly two approaches. The first approach (e.g., Libra, Citeseer, Cimple) employs semi-automatic methods to extract and integrate data from a multitude of data sources. The second approach (e.g., Wikipedia, Intellipedia) deploys an initial portal in wiki format, then invites community members to revise and add material. In this paper we consider combining the above two approaches to building community portals. The new hybrid machine-human approach brings significant benefits. It can achieve broader and deeper coverage, provide more incentives for users to contribute, and keep the portal more up-to-date with less user effort. In a sense, it enables building "community wikipedias", backed by an underlying structured database that is continuously updated using automatic techniques. We outline our ideas for the new approach, describe its challenges and opportunities, and provide initial solutions. Finally, we describe a real-world implementation and preliminary experiments that demonstrate the utility of the new approach.

We describe PNUTS, a massively parallel and geographically distributed database system for Yahoo!'s web applications. PNUTS provides data storage organized as hashed or ordered tables, low latency for large numbers of concurrent requests including updates and queries, and novel per-record consistency guarantees. It is a hosted, centrally managed, and geographically distributed service, and utilizes automated load-balancing and failover to reduce operational complexity. The first version of the system is currently serving in production. We describe the motivation for PNUTS and the design and implementation of its table storage and replication layers, and then present experimental results.

A fundamental concern of data integration in an XML context is the ability to embed one or more source documents in a target document so that (a) the target document conforms to a target schema and (b) the information in the source documents is preserved . In this paper, information preservation for XML is formally studied, and the results of this study guide the definition of a novel notion of schema embedding between two XML DTD schemas represented as graphs. Schema embedding generalizes the conventional notion of graph similarity by allowing an edge in a source DTD schema to be mapped to a path in the target DTD. Instance-level embeddings can be derived from the schema embedding in a straightforward manner, such that conformance to a target schema and information preservation are guaranteed. We show that it is NP-complete to find an embedding between two DTD schemas. We also outline efficient heuristic algorithms to find candidate embeddings, which have proved effective by our experimental study. These yield the first systematic and effective approach to finding information preserving XML mappings.

We propose a class of constraints, referred to as conditional functional dependencies (CFDs), and study their applications in data cleaning. In contrast to traditional functional dependencies (FDs) that were developed mainly for schema design, CFDs aim at capturing the consistency of data by incorporating bindings of semantic ally related values. For CFDs we provide an inference system analogous to Armstrong's axioms for FDs, as well as consistency analysis. Since CFDs allow data bindings, a large number of individual constraints may hold on a table, complicating detection of constraint violations. We develop techniques for detecting CFD violations in SQL as well as novel techniques for checking multiple constraints in a single query. We experimentally evaluate the performance of our CFD-based methods for inconsistency detection. This not only yields a constraint theory for CFDs but is also a step toward a practical constraint-based method for improving data quality.

This paper investigates a class of transform queries proposed by XQuery Update [6]. A transform query is defined in terms of XML update syntax. When posed on an XML tree T, it returns another XML tree that would be produced by executing its embedded update on T, without destructive impact on T. Transform queries support a variety of applications including XML hypothetical queries, the simulation of updates on virtual views, and the enforcement of XML access control. In light of the wide-range of applications for transform queries, we develop automaton-based techniques for efficiently evaluating transform queries and for computing their compositions with user queries in standard XQuery. We provide (a)three algorithms to implement transform queries without change to existing XQuery processors,(b) a linear-time algorithm, based on a seamless integration of automaton execution and SAX parsing, to evaluate transform queries on large XML documents that are difficult to handle by existing XQuery engines, and (c) an algorithm to rewrite the composition of user queries and transform queries into a single efficient query in standard XQuery. We also present experimental results comparing the efficiency of our evaluation and composition algorithms for transform queries.

Data integrated from multiple sources may contain inconsistencies that violate integrity constraints. The constraint repair problem attempts to find "low cost" changes that, when applied, will cause the constraints to be satisfied. While in most previous work repair cost is stated in terms of tuple insertions and deletions, we follow recent work to define a database repair as a set of value modifications. In this context, we introduce a novel cost framework that allows for the application of techniques from record-linkage to the search for good repairs. We prove that finding minimal-cost repairs in this model is NP-complete in the size of the database, and introduce an approach to heuristic repair-construction based on equivalence classes of attribute values. Following this approach, we define two greedy algorithms. While these simple algorithms take time cubic in the size of the database, we develop optimizations inspired by algorithms for duplicate-record detection that greatly improve scalability. We evaluate our framework and algorithms on synthetic and real data, and show that our proposed optimizations greatly improve performance at little or no cost in repair quality.

When large XML documents published from a database are maintained externally, it is inefficient to repeatedly recompute them when the database is updated. Vastly preferable is incremental update, as common for views stored in a data warehouse. However, to support schema-directed publishing, there may be no simple query that defines the mapping from the database to the external document. To meet the need for efficient incremental update, this paper studies two approaches for incremental evaluation of ATGs [4], a formalism for schema-directed XML publishing. The reduction approach seeks to push as much work as possible to the underlying DBMS. It is based on a relational encoding of XML trees and a nontrivial translation of ATGs to SQL 99 queries with recursion. However, a weakness of this approach is that it relies on high-end DBMS features rather than the lowest common denominator. In contrast, the bud-cut approach pushes only simple queries to the DBNS and performs the bulk of the work in middleware. It capitalizes on the tree-structure of XML views to minimize unnecessary recomputations and leverages optimization techniques developed for XML publishing. While implementation of the reduction approach is not yet in the reach of commercial DBMS, we have implemented the bud-cut approach and experimentally evaluated its performance compared to recomputation.

We present LegoDB, a cost-based XML storage mapping engine that automatically explores a space of possible XML-to-relational mappings and selects an efficient mapping for a given application.

No abstract available.

While the XML Stylesheet Language for Transformations (XSLT) was not designed as a query language, it is well-suited for many query-like operations on XML documents including selecting and restructuring data. Further, it actively fulfills the role of an XML query language in modern applications and is widely supported by application platform software. However, the use of database techniques to optimize and execute XSLT has only recently received attention in the research community. In this paper, we focus on the case where XSL transformations are to be run on XML documents defined as views of relational databases. For a subset of XSLT, we present an algorithm to compose a transformation with an XML view, eliminating the need for the XSLT execution. We then describe how to extend this algorithm to handle several additional features of XSLT, including a proposed approach for handling recursion. 1.

Increasingly, for extensibility and performance, special purpose application code is being integrated with database system code. Such application code has direct access to database system buffers, and as a result, the danger of data being corrupted due to inadvertent application writes is increased. Previously proposed hardware techniques to protect from corruption require system calls, and their performance depends on details of the hardware architecture. We investigate an alternative approach which uses codewords associated with regions of data to detect corruption and to prevent corrupted data from being used by subsequent transactions. We develop several such techniques which vary in the level of protection, space overhead, performance, and impact on concurrency. These techniques are implemented in the Dali main-memory storage manager, and the performance impact of each on normal processing is evaluated. Novel techniques are developed to recover when a transaction has read corrupted data caused by a bad write and gone on to write other data in the database. These techniques use limited and relatively low-cost logging of transaction reads to trace the corruption and may also prove useful when resolving problems caused by incorrect data entry and other logical errors.

As Web applications manipulate an increasing amount of XML, there is a growing interest in storing XML data in relational databases. Due to the mismatch between the complexity of XML's tree structure and the simplicity of flat relational tables, there are many ways to store the same document in an RDBMS, and a number of heuristic techniques have been proposed. These techniques typically define fixed mappings and do not take application characteristics into account. However, a fixed mapping is unlikely to work well for all possible applications. In contrast, LegoDB is a cost-based XML storage mapping engine that explores a space of possible XML-to-relational mappings and selects the best mapping for a given application. LegoDB leverages current XML and relational technologies: (1) it models the target application with an XML Schema, XML data statistics, and an XQuery workload; (2) the space of configurations is generated through XML-Schema rewritings; and (3) the best among the derived configurations is selected using cost estimates obtained through a standard relational optimizer. We describe the LegoDB storage engine and provide experimental results that demonstrate the effectiveness of this approach.

XML and other semi-structured data may have partially specified or missing schema information, motivating the use of a structural summary which can be automatically computed from the data. These summaries also serve as indices for evaluating the complex path expressions common to XML and semi-structured query languages. However, to answer all path queries accurately, summaries must encode information about long, seldom-queried paths, leading to increased size and complexity with little added value. We introduce the A(k)-indices, a family of approximate structural summaries. They are based on the concept of k-bisimilarity, in which nodes are grouped based on local structure, i.e., the incoming paths of length up to k. The parameter k thus smoothly varies the level of detail (and accuracy) of the A(k)-index. For small values of k, the size of the index is substantially reduced. While smaller, the A(k) index is approximate, and we describe techniques for efficiently extracting exact answers to regular path queries. Our experiments show that, for moderate values of k, path evaluation using the A(k)-index ranges from being very efficient for simple queries to competitive for most complex queries, while using significantly less space than comparable structures.

In this paper, we ask if the traditional relational query acceleration techniques of summary tables and covering indexes have analogs for branching path expression queries over tree- or graph-structured XML data. Our answer is yes --- the forward-and-backward index already proposed in the literature can be viewed as a structure analogous to a summary table or covering index. We also show that it is the smallest such index that covers all branching path expression queries. While this index is very general, our experiments show that it can be so large in practice as to offer little performance improvement over evaluating queries directly on the data. Likening the forward-and-backward index to a covering index on all the attributes of several tables, we devise an index definition scheme to restrict the class of branching path expressions being indexed. The resulting index structures are dramatically smaller and perform better than the full forward-and-backward index for these classes of branching path expressions. This is roughly analogous to the situation in multidimensional or OLAP workloads, in which more highly aggregated summary tables can service a smaller subset of queries but can do so at increased performance. We evaluate the performance of our indexes on both relational decompositions of XML and a native storage technique. As expected, the performance benefit of an index is maximized when the query matches the index definition.

The performance of main-memory index structures is increasingly determined by the number of CPU cache misses incurred when traversing the index. When keys are stored indirectly, as is standard in main-memory databases, the cost of key retrieval in terms of cache misses can dominate the cost of an index traversal. Yet it is inefficient in both time and space to store even moderate sized keys directly in index nodes. In this paper, we investigate the performance of tree structures suitable for OLTP workloads in the face of expensive cache misses and non-trivial key sizes. We propose two index structures, pkT-trees and pkB-trees, which significantly reduce cache misses by storing partial-key information in the index. We show that a small, fixed amount of key information allows most cache misses to be avoided, allowing for a simple node structure and efficient implementation. Finally, we study the performance and cache behavior of partial-key trees by comparing them with other main-memory tree structures for a wide variety of key sizes and key value distributions.

Increasingly, for extensibility and performance, special-purpose application code is being integrated with database system code. Such application code has direct access to database system buffers and, as a result, the danger of data being corrupted due to inadvertent application writes is increased. Previously proposed hardware techniques to protect data from corruption required system calls, and their performance depended on the details of the hardware architecture. We investigate an alternative approach which uses codewords associated with regions of data to detect corruption and to prevent corrupted data from being used by subsequent transactions. We develop several such techniques which vary in the level of protection, space overhead, performance and impact on concurrency. These techniques are implemented in the Dali/spl acute/ main-memory storage manager, and the performance impact of each on normal processing is evaluated. Novel techniques are developed to recover when a transaction has read corrupted data caused by a bad write, and then gone on to write other data in the database. These techniques use limited and relatively low-cost logging of transaction reads to trace the corruption, and may also prove useful when resolving problems caused by incorrect data entry and other logical errors.

No abstract available.