We demonstrate Pipemizer , an optimizer and recommender aimed at improving the performance of queries or jobs in pipelines. These job pipelines are ubiquitous in modern data analytics due to jobs reading output files written by other jobs. Given that more than 650k jobs run on Microsoft's SCOPE job service per day and about 70% have inter-job dependencies, identifying optimization opportunities across query jobs is of considerable interest to both cluster operators and users. Pipemizer addresses this need by providing recommendations to users, allowing users to understand their system, and facilitating automated application of recommendations. Pipemizer introduces novel optimizations that include holistic pipeline-aware statistics generation, inter-job operator push-up, and job split & merge. This demonstration showcases optimizations and recommendations generated by Pipemizer , enabling users to understand and optimize job pipelines.

Query processing over big data is ubiquitous in modern clouds, where the system takes care of picking both the physical query execution plans and the resources needed to run those plans, using a cost-based query optimizer. A good cost model, therefore, is akin to better resource efficiency and lower operational costs. Unfortunately, the production workloads at Microsoft show that costs are very complex to model for big data systems. In this work, we investigate two key questions: (i) can we learn accurate cost models for big data systems, and (ii) can we integrate the learned models within the query optimizer. To answer these, we make three core contributions. First, we exploit workload patterns to learn a large number of individual cost models and combine them to achieve high accuracy and coverage over a long period. Second, we propose extensions to Cascades framework to pick optimal resources, i.e, number of containers, during query planning. And third, we integrate the learned cost models within the Cascade-style query optimizer of SCOPE at Microsoft. We evaluate the resulting system, Cleo, in a production environment using both production and TPC-H workloads. Our results show that the learned cost models are 2 to 3 orders of magnitude more accurate, and 20X more correlated with the actual runtimes, with a large majority (70%) of the plan changes leading to substantial improvements in latency as well as resource usage.

Query optimizers are notorious for inaccurate cost estimates, leading to poor performance. The root of the problem lies in inaccurate cardinality estimates, i.e., the size of intermediate (and final) results in a query plan. These estimates also determine the resources consumed in modern shared cloud infrastructures. In this paper, we present C ARD L EARNER , a machine learning based approach to learn cardinality models from previous job executions and use them to predict the cardinalities in future jobs. The key intuition in our approach is that shared cloud workloads are often recurring and overlapping in nature, and so we could learn cardinality models for overlapping subgraph templates. We discuss various learning approaches and show how learning a large number of smaller models results in high accuracy and explainability. We further present an exploration technique to avoid learning bias by considering alternate join orders and learning cardinality models over them. We describe the feedback loop to apply the learned models back to future job executions. Finally, we show a detailed evaluation of our models (up to 5 orders of magnitude less error), query plans (60% applicability), performance (up to 100% faster, 3x fewer resources), and exploration (optimal in few 10s of executions).

Keyword search is a useful tool for exploring large RDF data sets. Existing techniques either rely on constructing a distance matrix for pruning the search space or building summaries from the RDF graphs for query processing. In this work, we show that existing techniques have serious limitations in dealing with realistic, large RDF data with tens of millions of triples. Furthermore, the existing summarization techniques may lead to incorrect/incomplete results. To address these issues, we propose an effective summarization algorithm to summarize the RDF data. Given a keyword query, the summaries lend significant pruning powers to exploratory keyword search and result in much better efficiency compared to previous works. Unlike existing techniques, our search algorithms always return correct results. Besides, the summaries we built can be updated incrementally and efficiently. Experiments on both benchmark and large real RDF data sets show that our techniques are scalable and efficient.

Temporal and multi-version databases are ideal candidates for a distributed store, which offers large storage space, and parallel and distributed processing power from a cluster of (commodity) machines. A key challenge is to achieve a good load balancing algorithm for storage and processing of these data, which is done by partitioning the database. We introduce the concept of optimal splitters for temporal and multi-version databases, which induce a partition of the input data set, and guarantee that the size of the maximum bucket be minimized among all possible configurations, given a budget for the desired number of buckets. We design efficient methods for memory- and disk resident data respectively, and show that they significantly outperform competing baseline methods both theoretically and empirically on large real data sets.

This paper revisits the classical problem of multi-query optimization in the context of RDF/SPARQL. We show that the techniques developed for relational and semi-structured data/query languages are hard, if not impossible, to be extended to account for RDF data model and graph query patterns expressed in SPARQL. In light of the NP-hardness of the multi-query optimization for SPARQL, we propose heuristic algorithms that partition the input batch of queries into groups such that each group of queries can be optimized together. An essential component of the optimization incorporates an efficient algorithm to discover the common sub-structures of multiple SPARQL queries and an effective cost model to compare candidate execution plans. Since our optimization techniques do not make any assumption about the underlying SPARQL query engine, they have the advantage of being portable across different RDF stores. The extensive experimental studies, performed on three popular RDF stores, show that the proposed techniques are effective, efficient and scalable.

Query execution assurance is an important concept in defeating lazy servers in the database as a service model. We show that extending query execution assurance to outsourced databases with multiple data owners is highly inefficient. To cope with lazy servers in the distributed setting, we propose query access assurance (Qaa) that focuses on IO-bound queries. The goal in Qaa is to enable clients to verify that the server has honestly accessed all records that are necessary to compute the correct query answer, thus eliminating the incentives for the server to be lazy if the query cost is dominated by the IO cost in accessing these records. We formalize this concept for distributed databases, and present two efficient schemes that achieve Qaa with high success probabilities. The first scheme is simple to implement and deploy, but may incur excessive server to client communication cost and verification cost at the client side, when the query selectivity or the database size increases. The second scheme is more involved, but successfully addresses the limitation of the first scheme. Our design employs a few number theory techniques. Extensive experiments demonstrate the efficiency, effectiveness, and usefulness of our schemes.

The problem of answering SPARQL queries over virtual SPARQL views is commonly encountered in a number of settings, including while enforcing security policies to access RDF data, or when integrating RDF data from disparate sources. We approach this problem by rewriting SPARQL queries over the views to equivalent queries over the underlying RDF data, thus avoiding the costs entailed by view materialization and maintenance. We show that SPARQL query rewriting combines the most challenging aspects of rewriting for the relational and XML cases: like the relational case, SPARQL query rewriting requires synthesizing multiple views; like the XML case, the size of the rewritten query is exponential to the size of the query and the views. In this paper, we present the first native query rewriting algorithm for SPARQL. For an input SPARQL query over a set of virtual SPARQL views, the rewritten query resembles a union of conjunctive queries and can be of exponential size. We propose optimizations over the basic rewriting algorithm to (i) minimize each conjunctive query in the union; (ii) eliminate conjunctive queries with empty results from evaluation; and (iii) efficiently prune out big portions of the search space of empty rewritings. The experiments, performed on two RDF stores, show that our algorithms are scalable and independent of the underlying RDF stores. Furthermore, our optimizations have order of magnitude improvements over the basic rewriting algorithm in both the rewriting size and evaluation time.