Analysis and Modeling of Social Influence in High Performance Computing Workloads

TL;DR

This paper investigates social influence in HPC workloads, revealing social patterns similar to social networks, and proposes an efficient online algorithm to identify and track social groups affecting workload behavior.

Contribution

It introduces a novel analysis of social influence in HPC workloads and develops a fast, online learning algorithm for identifying and monitoring social groups.

Findings

Social influence creates bursty HPC workload behavior.

A power-law distribution characterizes the social graph among HPC users.

The proposed online algorithm efficiently identifies social groups with minimal data.

Abstract

Social influence among users (e.g., collaboration on a project) creates bursty behavior in the underlying high performance computing (HPC) workloads. Using representative HPC and cluster workload logs, this paper identifies, analyzes, and quantifies the level of social influence across HPC users. We show the existence of a social graph that is characterized by a pattern of dominant users and followers. This pattern also follows a power-law distribution, which is consistent with those observed in mainstream social networks. Given its potential impact on HPC workloads prediction and scheduling, we propose a fast-converging, computationally-efficient online learning algorithm for identifying social groups. Extensive evaluation shows that our online algorithm can (1) quickly identify the social relationships by using a small portion of incoming jobs and (2) can efficiently track group evolution over time.