Wednesday, 13 December 2017

Session 3: Research Track Best Paper Candidates

Towards Energy-Proportional Computing for Enterprise-Class Server Workloads

Authors:

Balaji Subramaniam (Virginia Tech)
Wu-chun Feng (Virginia Tech)

Abstract:

Massive data centers housing thousands of computing nodes have become commonplace in enterprise computing, and the power consumption of such data centers is growing at an unprecedented rate. Adding to the problem is the inability of the servers to exhibit energy proportionality, i.e., provide energy-efficient execution under all levels of utilization, which diminishes the overall energy efficiency of the data center. It is imperative that we realize effective strategies to control the power consumption of the server and improve the energy efficiency of data centers. With the advent of Intel Sandy Bridge processors, we have the ability to specify a limit on power consumption during runtime, which creates opportunities to design new power-management techniques for enterprise workloads and make the systems that they run on more energy proportional.

In this paper, we investigate whether it is possible to achieve energy proportionality for an enterprise-class server workload, namely SPECpower ssj2008 benchmark, by using Intel’s Running Average Power Limit (RAPL) interfaces. First, we analyze the power consumption and characterize the instantaneous power profile of the SPECpower benchmark within different subsystems using the on-chip energy meters exposed via the RAPL interfaces. We then analyze the impact of RAPL power limiting on the performance, per-transaction response time, power consumption, and energy efficiency of the benchmark under different load levels. Our observations and results shed light on the efficacy of the RAPL interfaces and provide guidance for designing power-management techniques for enterprise-class workloads.

DOI: 10.1145/2479871.2479878

Full text: PDF

[#][]

Automated Root Cause Isolation of Performance Regressions During Software Development

Authors:

Christoph Heger (Karlsruhe Institute of Technology)
Jens Happe (SAP Research)
Roozbeh Farahbod (SAP Research)

Abstract:

Performance is crucial for the success of an application. To build responsive and cost efficient applications, software engineers must be able to detect and fix performance problems early in the development process. Existing approaches are either relying on a high level of abstraction such that critical problems cannot be detected or require high manual effort. In this paper, we present a novel approach that integrates performance regression root cause analysis into the existing development infrastructure using performance-aware unit tests and the revision history. Our approach is easy to use and provides software engineers immediate insights with automated root cause analysis. In a realistic case study based on the change history of Apache Commons Math, we demonstrate that our approach can automatically detect and identify the root cause of a major performance regression.

DOI: 10.1145/2479871.2479879

Full text: PDF

[#][]

Multiple Class G-Networks with Restart

Authors:

Jean-Michel Fourneau (Université de Versailles St Quentin)
Katinka Wolter (Freie Universität Berlin)
Philipp Reinecke (HP Labs Bristol)
Tilman Kraufl (Freie Universität Berlin)
Alexandra Danilkina (Freie Universität Berlin)

Abstract:

Restart is a common technique for improving response-times in complex systems where the causes of delays can either not be discerned, or not be addressed by the user. With restart, the user aborts a running job that exceeds a deadline, and resubmits it to the system immediately. In many common scenarios, this approach can reduce the response-times that the user experiences. Restart has been well-studied for scenarios where only one user applies restart, and typically in cases where queueing effects can be neglected. In this paper we approach the question of restart in a scenario where restart is applied by many users in a system that can be modelled as an open queueing network. We apply the G-Networks formalism to this problem. We use negative customers to model the abortion and retry of a request. The open G-network uses multiple classes with phase-type distributed service times. This allows the approximation of a preemptive repeat different behaviour as it is natural for multiple restarts of a request. We compute the response time of a request and show that an optimal restart interval can be found. The results are compared with simulation.

DOI: 10.1145/2479871.2479880

Full text: PDF

[#][]