Filter before you parse

Abstract

Exploratory big data applications often run on raw unstructured or semi-structured data formats, such as JSON files or text logs. These applications can spend 80-90% of their execution time parsing the data. In this paper, we propose a new approach for reducing this overhead: apply filters on the data's raw bytestream before parsing. This technique, which we call raw filtering, leverages the features of modern hardware and the high selectivity of queries found in many exploratory applications. With raw filtering, a user-specified query predicate is compiled into a set of filtering primitives called raw filters (RFs). RFs are fast, SIMD-based operators that occasionally yield false positives, but never false negatives. We combine multiple RFs into an RF cascade to decrease the false positive rate and maximize parsing throughput. Because the best RF cascade is datadependent, we propose an optimizer that dynamically selects the combination of RFs with the best expected throughput, achieving within 10% of the global optimum cascade while adding less than 1.2% overhead. We implement these techniques in a system called Sparser, which automatically manages a parsing cascade given a data stream in a supported format (e.g., JSON, Avro, Parquet) and a user query. We show that many real-world applications are highly selective and benefit from Sparser. Across diverse workloads, Sparser accelerates state-of-the-art parsers such as Mison by up to 22× and improves end-to-end application performance by up to 9×.

The Case For Alternative Web Archival Formats To Expedite The Data-To-Insight Cycle

“…This conjecture is corroborated by similar applications analyzing large JSON files or text logs. Two recent research studies estimated that they spend 80-90% of the execution time on parsing [32,46]. In comparison, Freeman et al [21] put this estimate to less than 40-70% when analyzing time-series data in binary encoding.…”

“…Two standard data formats, Parquet [6] and Avro [7], have been developed from these projects and adopted as their default data persistence formats. Although not intended for this comparison, circumstantial evidence in [46] indicates that parsing data formatted in Parquet and Avro can be at least one order of magnitude faster than parsing the same data formatted in JSON.…”

“…Similarly, gap-filing software have been developed to improve the I/O performance for using HDF5, netCDF [35,58] or FITS [48] in a Spark environment. Further, many have worked towards improving the parsing performance of the inefficient yet widely adopted JSON format [32,46]. These improvements leverage various hardware/software performance gaps [59] but can not fully eliminate the inherent inefficiency incurred by a low-performing data format.…”

JSON Tiles: Fast Analytics on Semi-Structured Data

“…For both structured and semi-structured data, parsers such as FAD.js, Mison or SIMD-JSON use modern CPU properties for fast reads [12,27,37,39]. Raw filters are used to speed up the parsing and reduce the amount of data ingested into the database [48,63].…”

What You Need to Know About (Smart) Network Interface Cards

“…Such NICs are referred to as SmartNICs. Several preliminary investigations of SmartNIC technologies have demonstrated potential benefits for offloading networking stacks [2,10,30,31,32], network functions [3,18,25,43,4], key-value stores [7,26,28], packet schedulers [44], neural networks [42], and beyond [21,38]. Despite the increasing relevance of (smart) NICs in today's systems, very few studies have focused on dissecting the performance of SmartNICs, comparing them with their predecessors, and providing guidelines for deploying NIC-offloaded applications, with a focus on packet classification.…”