Matching Mechanisms For Real-Time Computational Resource Exchange Markets

TL;DR

This paper introduces matching mechanisms for a real-time computational resource exchange market called Chital, aiming to incentivize peer computation, discourage dishonesty, and optimize overall performance through novel algorithms and theoretical analysis.

Contribution

It presents new matching algorithms for resource exchange markets, evaluates their performance, and derives a near-optimal algorithm under specific assumptions.

Findings

The proposed algorithm nearly always benefits users compared to solo computation.

Empirical evaluation shows the effectiveness of the matching algorithms in simulated environments.

A theoretically near-optimal algorithm is derived under certain distributional assumptions.

Abstract

In this paper we describe matching mechanisms for a real-time computational resource exchange market, Chital, that incentivizes participating clients to perform computation for their peers in exchange for overall improved performance. The system is designed to discourage dishonest behavior via a credit system, while simultaneously minimizing the use of dedicated computing servers and the number of verifications performed by the administrating servers. We describe the system in the context of a pre-existing system (under development), Vedalia \cite{715Project}, for analyzing and visualizing product reviews, by using machine learning such as topic models. We extend this context to general computing tasks, describe a list of matching algorithms, and evaluate their performance in a simulated environment. In addition, we design a matching algorithm that optimizes the amount of time a participant could save compared to computing a task on their own, and show empirically that this algorithm results in a situation in which it is almost always optimal for a user to join the exchange than do computation alone. Lastly, we use a top-down approach to derive a theoretically near-optimal matching algorithm under certain distributional assumptions on query frequency.