Scalable Social Coordination using Enmeshed Queries

TL;DR

This paper introduces enmeshed queries, a flexible model for social coordination that supports declarative group formation based on social attributes, and presents efficient heuristics for scalable implementation.

Contribution

The paper proposes enmeshed queries as a novel, general model for social coordination and develops heuristics for efficient, scalable matching algorithms.

Findings

Heuristic algorithms achieve 86% of optimal matchings.

Average query processing time is 40 microseconds.

Algorithm scales well with multiple cores and servers.

Abstract

Social coordination allows users to move beyond awareness of their friends to efficiently coordinating physical activities with others. While specific forms of social coordination can be seen in tools such as Evite, Meetup and Groupon, we introduce a more general model using what we call enmeshed queries. An enmeshed query allows users to declaratively specify an intent to coordinate by specifying social attributes such as the desired group size and who/what/when, and the database returns matching queries. Enmeshed queries are continuous, but new queries (and not data) answer older queries; the variable group size also makes enmeshed queries different from entangled queries, publish-subscribe systems, and dating services. We show that even offline group coordination using enmeshed queries is NP-hard. We then introduce efficient heuristics that use selective indices such as location and time to reduce the space of possible matches; we also add refinements such as delayed evaluation and using the relative matchability of users to determine search order. We describe a centralized implementation and evaluate its performance against an optimal algorithm. We show that the combination of not stopping prematurely (after finding a match) and delayed evaluation results in an algorithm that finds 86% of the matches found by an optimal algorithm, and takes an average of 40 usec per query using 1 core of a 2.5 Ghz server machine. Further, the algorithm has good latency, is reasonably fair to large group size requests, and can be scaled to global workloads using multiple cores and multiple servers. We conclude by describing potential generalizations that add prices, recommendations, and data mining to basic enmeshed queries.