The X-Files: Investigating Alien Performance in a Thin-client World

TL;DR

This paper investigates performance issues in X11-based seismic applications on a large cluster, identifying key problems like long launch times and poor responsiveness, and demonstrates workload analysis as a diagnostic approach.

Contribution

It provides a detailed case study of performance bottlenecks in X11 applications, highlighting diagnostic methods for identifying UFOs in a high-performance computing environment.

Findings

Long application launch times observed

Poor interactive response times documented

Workload analysis used to diagnose UFOs

Abstract

Many scientific applications use the X11 window environment; an open source windows GUI standard employing a client/server architecture. X11 promotes: distributed computing, thin-client functionality, cheap desktop displays, compatibility with heterogeneous servers, remote services and administration, and greater maturity than newer web technologies. This paper details the author's investigations into close encounters with alien performance in X11-based seismic applications running on a 200-node cluster, backed by 2 TB of mass storage. End-users cited two significant UFOs (Unidentified Faulty Operations) i) long application launch times and ii) poor interactive response times. The paper is divided into three major sections describing Close Encounters of the 1st Kind: citings of UFO experiences, the 2nd Kind: recording evidence of a UFO, and the 3rd Kind: contact and analysis. UFOs do exist and this investigation presents a real case study for evaluating workload analysis and other diagnostic tools.