The Marriage of Incremental and Approximate Computing

TL;DR

This paper introduces IncApprox, a system that combines incremental and approximate computing by biasing sampling towards memoized data, enabling efficient, low-latency data analytics with bounded error.

Contribution

It proposes a novel online stratified sampling algorithm that marries incremental and approximate computing paradigms, implemented in Apache Spark Streaming.

Findings

IncApprox achieves low-latency analytics with bounded error.

The system effectively combines benefits of incremental and approximate computing.

Experimental results show improved efficiency over traditional methods.

Abstract

Most data analytics systems that require low-latency execution and efficient utilization of computing resources, increasingly adopt two computational paradigms, namely, incremental and approximate computing. Incremental computation updates the output incrementally instead of re-computing everything from scratch for successive runs of a job with input changes. Approximate computation returns an approximate output for a job instead of the exact output. Both paradigms rely on computing over a subset of data items instead of computing over the entire dataset, but they differ in their means for skipping parts of the computation. Incremental computing relies on the memoization of intermediate results of sub-computations, and reusing these memoized results across jobs for sub-computations that are unaffected by the changed input. Approximate computing relies on representative sampling of the entire dataset to compute over a subset of data items. In this thesis, we make the observation that these two computing paradigms are complementary, and can be married together! The high level idea is to: design a sampling algorithm that biases the sample selection to the memoized data items from previous runs. To concretize this idea, we designed an online stratified sampling algorithm that uses self-adjusting computation to produce an incrementally updated approximate output with bounded error. We implemented our algorithm in a data analytics system called IncAppox based on Apache Spark Streaming. Our evaluation of the system shows that IncApprox achieves the benefits of both incremental and approximate computing.