[Paper Review] OLA-RAW: Scalable Exploration over Raw Data

Yu Cheng, Weijie Zhao+1 more|arXiv (Cornell University)|Feb 1, 2017

Advanced Database Systems and Queries36 references3 citations

TL;DR

OLA-RAW introduces a query-driven, bi-level sampling scheme for parallel online aggregation over raw data, enabling fast, accurate estimation without data reorganization. By incrementally building a memory-resident sample synopsis during in-situ query execution, it reduces I/O and CPU costs, achieving up to 10X performance improvement over external tables and chunk-level sampling while ensuring accurate early termination for uninteresting queries.

ABSTRACT

In-situ processing has been proposed as a novel data exploration solution in many domains generating massive amounts of raw data, e.g., astronomy, since it provides immediate SQL querying over raw files. The performance of in-situ processing across a query workload is, however, limited by the speed of full scan, tokenizing, and parsing of the entire data. Online aggregation (OLA) has been introduced as an efficient method for data exploration that identifies uninteresting patterns faster by continuously estimating the result of a computation during the actual processing---the computation can be stopped as early as the estimate is accurate enough to be deemed uninteresting. However, existing OLA solutions have a high upfront cost of randomly shuffling and/or sampling the data. In this paper, we present OLA-RAW, a bi-level sampling scheme for parallel online aggregation over raw data. Sampling in OLA-RAW is query-driven and performed exclusively in-situ during the runtime query execution, without data reorganization. This is realized by a novel resource-aware bi-level sampling algorithm that processes data in random chunks concurrently and determines adaptively the number of sampled tuples inside a chunk. In order to avoid the cost of repetitive conversion from raw data, OLA-RAW builds and maintains a memory-resident bi-level sample synopsis incrementally. We implement OLA-RAW inside a modern in-situ data processing system and evaluate its performance across several real and synthetic datasets and file formats. Our results show that OLA-RAW chooses the sampling plan that minimizes the execution time and guarantees the required accuracy for each query in a given workload. The end result is a focused data exploration process that avoids unnecessary work and discards uninteresting data.

Motivation & Objective

• To address the inefficiency of traditional in-situ data processing, which requires full scans and incurs high I/O and CPU costs even for uninteresting queries.
• To overcome the limitations of existing online aggregation (OLA) systems, which suffer from high upfront costs due to data shuffling or sampling before query execution.
• To enable early termination of uninteresting queries by providing accurate, confidence-bound aggregate estimates during runtime without data reorganization.
• To design a scalable, parallel online aggregation framework that incrementally maintains a bi-level sample synopsis directly from raw data files.
• To minimize execution time while guaranteeing required accuracy for each query in a workload, focusing exploration on potentially interesting data.

Proposed method

• OLA-RAW employs a novel resource-aware bi-level sampling algorithm that processes data in random chunks in parallel during query execution.
• It adaptively determines the number of sampled tuples within each chunk based on resource availability and query requirements.
• The system incrementally builds and maintains a memory-resident bi-level sample synopsis to avoid repeated conversion from raw data formats.
• Sampling is performed exclusively in-situ during query execution, eliminating the need for data reorganization or pre-processing.
• The method uses a bi-level Bernoulli sampling strategy that avoids the inspection paradox and supports accurate confidence bounds for aggregate estimations.
• OLA-RAW integrates with a modern in-situ data processing system to support SQL queries directly over raw files, such as FITS and other flat-file formats.

Experimental results

Research questions

• RQ1 Can online aggregation over raw data be made scalable and efficient without requiring data reorganization or upfront sampling?
• RQ2 How can bi-level sampling be adapted to in-situ query execution to minimize I/O and CPU overhead?
• RQ3 Can a memory-resident sample synopsis be incrementally maintained during query execution to avoid repeated parsing of raw data?
• RQ4 What is the performance gain of query-driven, in-situ bi-level sampling compared to traditional external tables and chunk-level sampling?
• RQ5 To what extent can OLA-RAW enable early termination of uninteresting queries while maintaining required accuracy?

Key findings

• OLA-RAW reduces execution time by up to 10X compared to external tables and chunk-level sampling across real and synthetic datasets.
• The system achieves accurate aggregate estimation with confidence bounds that allow correct early termination when results are uninteresting.
• By maintaining a memory-resident bi-level sample synopsis incrementally, OLA-RAW avoids the cost of repeated data conversion from raw formats.
• The query-driven sampling strategy enables focused data exploration, discarding uninteresting data early and reducing unnecessary computation.
• OLA-RAW outperforms existing OLA solutions by eliminating expensive upfront data shuffling or sampling, making it suitable for large-scale raw data exploration.
• The approach is effective across diverse file formats, including FITS, and scales well in parallel query execution environments.