OPQL: پرس و جوی منشأ جریان کاری علمی در سطح نمودار
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|22017||2013||23 صفحه PDF||سفارش دهید||محاسبه نشده|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Data & Knowledge Engineering, Volume 88, November 2013, Pages 37–59
Provenance has become increasingly important in scientific workflows to understand, verify, and reproduce the result of scientific data analysis. Most existing systems store provenance data in provenance stores with proprietary provenance data models and conduct query processing over the physical provenance storages using query languages, such as SQL, SPARQL, and XQuery, which are closely coupled to the underlying storage strategies. Querying provenance at such low level leads to poor usability of the system: a user needs to know the underlying schema to formulate queries; if the schema changes, queries need to be reformulated; and queries formulated for one system will not run in another system. In this paper, we present OPQL, a provenance query language that enables the querying of provenance directly at the graph level. An OPQL query takes a provenance graph as input and produces another provenance graph as output. Therefore, OPQL queries are not tightly coupled to the underlying provenance storage strategies. Our main contributions are: (i) we design OPQL, including six types of graph patterns, a provenance graph algebra, and OPQL syntax and semantics, that supports querying provenance at the graph level; (ii) we implement OPQL using a Web service via our OPMProv system; therefore, users can invoke the Web service to execute OPQL queries in a provenance browser, called OPMProVis. The result of OPQL queries is displayed as a provenance graph in OPMProVis. An experimental study is conducted to evaluate the feasibility and performance of OPMProv on OPQL provenance querying.
Provenance, which is one kind of metadata that captures the derivation history of a data product, including its original data sources, intermediate products, and the steps that were applied to produce it, has become increasingly important in the area of scientific workflows , , , , , ,  and  to interpret, validate, and analyze the result of scientific computing. In general, provenance captures past workflow execution and data derivation information (i.e., which tasks were performed and how data products were derived) via a provenance collection mechanism during workflow execution. Provenance captured typically holds data dependencies, process dependencies, causality between data and processes, and annotations. Such provenance is often represented by a provenance graph. For example, Fig. 1(a) shows a sample scientific workflow (which is the Load Workflow defined in the Third Provenance Challenge ) that checks and reads CSV files before loading, creates a database to load CSV files, loads them into tables and validates tables, and compacts a database after loading. Fig. 1(b) shows a sample provenance graph produced via the execution of the Load Workflow, where a node of a rectangle shape represents a process (i.e., a task), a node of an ellipse shape represents an artifact (i.e., a data product), which was used or generated by a process, a node of an octagon shape represents a contextual entity acting as a catalyst of a process, and an edge represents a causal dependency between its source denoting the effect and its destination denoting the cause. Fig. 1(c) and (d) also represents a data dependency graph associated with artifact a9 and a process dependency graph associated with process p8 from the provenance graph, respectively.Most existing systems , ,  and  store provenance data in provenance stores with proprietary provenance data models and conduct provenance querying using languages such as SQL, SPARQL, and XQuery, over the physical provenance storages (i.e., relational, RDF, and XML). Such query languages are closely coupled to the underlying provenance storage strategies. As a result, querying provenance at such a low level leads to poor usability of the system because users need to know the underlying storage schema to formulate queries; if the schema changes, queries need to be reformulated; and queries formulated for one system will not run in another system. Moreover, to formulate complicated provenance queries, a user requires the expertise about grammars, syntax, and semantics of a query language. Using existing approaches, provenance lineage queries (queries for tracking ancestor nodes) often require a user to write recursive queries, directly typing recursive statements or using recursive functionality. For example, Fig. 2 shows two query languages SQL and OPQL answering for a provenance query (Q1), which asks for “which artifacts contributed to derive artifact a5” over the provenance graph in Fig. 1(b). First, the SQL statement is expressed by a recursive query via the WITH ~ UNION ALL clause to track ancestor nodes associated with artifact a5; thus, to answer query Q1, a user has to know the information of table WasDerivedFrom (i.e., how attributes define and what the attributes mean), a user needs the expertise to formulate a recursive query, which is nontrivial, and if the schema (i.e., the table information) changes, the SQL query needs to be reformulated, which is cumbersome. On the other hand, since the OPQL query is formulated at the graph level, a user does not have to know the storage schema; therefore, even though the storage schema changes, it does not affect the query. Moreover, OPQL construct WDF supports a recursive query pattern; thus, OPQL is easy and convenient to formulate a recursive query pattern.In this paper, to address these issues, we propose OPQL, a provenance query language that enables the querying of provenance directly at the graph level. OPQL relies on the Open Provenance Model , a community-driven data model, which captures main aspects of the workflow provenance and does not enforce a particular physical representation of the provenance data. An OPQL query takes an OPM-compliant provenance graph as input and produces another OPM-compliant provenance graph as output. Thus, OPQL queries are not tightly coupled to the underlying storage strategies. This paper has the following main contributions: (1) we design OPQL, including six types of graph patterns, a provenance graph algebra, and OPQL syntax and semantics, that supports querying provenance at the graph level; and (2) we implement OPQL using a Web service via the OPMProv system; therefore, users can invoke the Web service to execute OPQL queries in a provenance browser, called OPMProVis. The result of OPQL queries is displayed as a provenance graph in OPMProVis. The rest of the paper is organized as follows. In Section 2, we overview the OPMProv system that stores, reasons, and queries scientific workflow provenance using a relational database. In Section 3, we present six types of graph patterns, a provenance graph algebra, and OPQL syntax and semantics to support querying provenance at the graph level. In Section 4, we discuss how OPQL is implemented via our OPMProv system. Section 5 describes how OPM graphs are reconstructed using our GraphConstruct algorithm. Section 6 reports the experimental study confirming the feasibility of using OPQL to query provenance at the graph level. Section 7 discusses related work on provenance query processing in existing systems. Section 8 presents conclusions and discusses possible future research directions. Finally, Acknowledgments section concludes the paper.
نتیجه گیری انگلیسی
In this paper, we designed the OPQL query language, including six types of graph patterns, a provenance graph algebra, and OPQL syntax and semantics, that enables querying of provenance at the graph level. We then implemented OPQL using a Web service via our OPMProv system; therefore, users can invoke the Web service to execute OPQL queries in user-friendly GUIs, such as OPMProVisD and OPMProVisW. Finally, we conducted experiments to evaluate the performance and feasibility of OPMProv on OPQL provenance querying, and the experimental results showed reasonable performance. In the future we plan to continue our research with OPMProv in four major directions. First, we plan to benchmark the performance of OPMProv using tools such, as the University of Texas Provenance Benchmark , as well as to compare the querying performance of our system with other provenance querying systems. Second, while in this paper we focused on querying a single provenance graph, in the future we will explore querying multiple provenance graphs. For example, if several workflow runs used the same artifact as a workflow input, the need may arise to find all artifacts derived from this artifact across provenance graphs representing several workflow runs. Third, we plan to explore query optimization , in which among other things we will study the performance of all 14 queries from the Third Provenance Challenge. Finally, we plan to study provenance security , particularly compliance management. This may include, for example managing permissions to read, write and modify artifacts, as well as history of reads, writes and modifications. While much research has been done on database usability, usability research in provenance querying is still in its infancy . This paper takes one of the first steps to query provenance at the graph level. More study about usability, especially from an end user perspective is needed in the future