1. ½ day: Energy Efficiency in High-Performance and Embedded Computing
2. ½ day: Benchmarking and Measuring Approximate Computing
1. Full-Day: Interactive Tutorial: Building Online Power Models from Real Data
2. ½ day Representative Datacenter workloads and experiments
3. ½ day: Programming Distributed Shared Memory Using ArgoDSM and the SPLASH-3 benchmarks
4. ½ day: Using Container to Uniformly Package, Distribute, and Execute Scientific Applications on Multiple Platforms
Title: Energy Efficiency in High-Performance and Embedded Computing
Organisers: Adept project (www.adept-project.eu)
Description: The Adept project addresses the challenge of energy-efficient use of parallel technologies. Adept builds on the expertise of software developers from high- performance computing (HPC) to exploit parallelism for performance, and on the expertise of Embedded systems engineers in managing energy usage. The aim of Adept is to develop a tool that will guide software developers and help them to model and predict the power consumption and performance of parallel software and hardware.
In line with the project's goals this workshop will bring together experts from the HPC and Embedded domains to discuss the current state-of-the-art and research in the area of energy-efficiency in parallel programming, with a focus on the measurement and prediction of energy usage. Specific emphasis will also be given to the cross-fertilisation of expertise between the two domains. The workshop will present the current state-of- the-art in measuring and modelling the power/performance of parallel software and hardware.
Workshop format and duration: The workshop is planned for a half-day. The format of the workshop will be a series talks from the Adept project, giving details on the project's objectives and outcomes.
Title: Benchmarking and Measuring Approximate Computing
Organisers: Georgios Karakonstantis (Queen's University Belfast, UK), Nikolaos Bellas (University of Thessaly, GR), Dimitris Gizopoulos (University of Athens, GR)
Description: Research in the last few years has focused on approximate computing as a means to overcome energy scaling barriers and reliability challenges of computer systems by utilizing the inherent error resilience of algorithms in various application domains such as multimedia and data analytics.
However, research in approximate computing is difficult to be evaluated quantitatively due to the lack of well-understood metrics that combine the elusive concept of application quality with well-understood metrics for performance and power dissipation. Moreover, the lack of a benchmark suite, comprising both kernels and complete applications, serving as a single point of reference for comparing the merit of different approaches is another obstacle that makes approximate computing research fragmented and ad-hoc.
This full day workshop will bring together researchers and practitioners actively working in this area with the aim of defining metrics for approximate computing and assembling a relevant benchmark suite. The program of the workshop consists of invited talks and discussion panels and is open to everyone that wants to participate.
Workshop format and duration: The workshop is planned for a half-day. The format of the workshop will be a series talks from the invited speakers and a panel discussion open to the participants.
Title: Building Online Power Models from Real Data
Organisers/Presenters: Matt Walker (University of Southampton, UK), Vasileios Spiliopoulos (Uppsala University, Sweden), Stephan Diestelhorst (Staff Research Engineer at ARM, Cambridge, UK)
Description: An interactive tutorial that explains methodologies to model the power consumption of modern mobile SoCs with a combination of tutorial presentations by domain experts, practical measurement experiments on real devices, and a fun competition between groups of participants to get the best power model of an SoC in an experimentation board supplied by us.
Duration: Full-Day Interactive Tutorial
Title: Representative Datacenter workloads and experiments
Organisers/Presenters: Ahmed Aley, Umea University (Umea university)
Description: There are multiple cloud benchmarks available today, including, CloudSuite, C-Mart, SPEC-Cloud benchmark, and PerfKit Benchmarker. This tutorial will cover a number of these benchmarks in some details covering the pros and cons of each of them, with hands-on experiments to show these pros and cons. We will then discuss how researchers in academia with no access to large-scale datacenter workloads use open-source software, and available data to replicate large-scale datacenter workloads. In particular, we will show participants how to replicate a Wikipedia installation with replayed real traces to perform experiments that are representative of a real application and a real workload running in a datacenter.
1. Introduction to Cloud benchmarks (~45 minutes)
2. Hands-on session using a few available benchmarks (~120 minutes)
3. Introduction to Wikipedia, MediaWiki, and the available workloads (~75 minutes)
4. Replicating a Wikipedia installation (~90 minutes)
Duration: Half a day on Sunday
Title: Programming Distributed Shared Memory Using ArgoDSM and the SPLASH-3 benchmarks
Organisers/Presenters: Magnus Norgren, David Klaftenegger, and Chris Sakalis (Uppsala University)
Description: This tutorial builds explains programming on the ArgoDSM introduced in HPDC 2015 (FCRC). The aim is to showcase the kind of programming ArgoDSM requires, and enables.
Duration: Half a day on Sunday
Title: Using Container to Uniformly Package, Distribute, and Execute Scientific Applications on Multiple Platforms
Organisers/Presenters: Alessandro Ferreira Leite, Institute for Technological Research SystemX and University of Brasilia
Description: Docker is an open source engine for deploying and running applications on Linux containers. Linux containers rely on kernel's features such as namespaces, cgroups, among others to sandbox processes into configurable virtual environments. Thanks to the Docker, Linux container has gained considerable attention over the recent years. Thus, it may be difficult to figure out what Linux containers and Docker are, and what they can do for us. Therefore, this tutorial will provide for attendees an end-to-end understanding of Linux containers and how to operate and to orchestrate them. They will learn the concepts that make up Linux containers, as well as the best practices on dealing with containers, and some advanced containers concepts.
This tutorial is intended for everyone interested in Linux containers, DevOps, and software infrastructure management. It does not assume any prior knowledge of Linux containers, but the participants must be comfortable with command line and the usage of a text editor. Moreover, it demands: (a) a laptop running Linux, Mac, or Windows operating system; (b) the VirtualBox 5.01 and the VM VirtualBox Extension Pack2 installed; (c) an SSH client (e.g., Putty3 for Windows users); (d) at least 15GB of free space for running the virtual machine images; and (e) Vagrant4 installed on the computer to manage the virtual machines.
Duration: Half day on Sunday (4 hours with a break of 15 minutes)