Tutorial Proposal for IEEE Cluster 2012

Title of Tutorial:

 Designing Clusters with InfiniBand and High-Speed Ethernet:Current Status and Trends

Type of the Tutorial:

Half-day Tutorial

Abstract:

InfiniBand (IB) and High-Speed Ethernet (HSE) interconnects are generating a lot of excitement towards building next generation High Performance Computing (HPC) clusters and enterprise datacenters. This tutorial will provide an overview of these emerging interconnects, the features they offer, their current market standing, and their suitability for cluster computing. It will start with a brief overview of IB, HSE and their architectural features. An overview of the emerging OpenFabrics stack which encapsulates both IB and Ethernet in a unified manner, and hardware technologies such as Virtual Protocol Interconnect (VPI), RDMA over Converged Enhanced Ethernet (RoCE) that aim at converged hardware solutions will be presented. IB and HSE hardware/software solutions and the market trends will be highlighted. Finally, sample performance numbers highlighting the performance these technologies can achieve in different environments such as MPI, PGAS/UPC, Parallel File Systems, Memcached and Cloud Computing (Hadoop, HDFS and HBase) will be shown.

Scope and Relevance:

Modern network interconnects such as IB and HSE have many novel features which were not available in previous networks. Apart from raw performance (8-, 16-, 24-, 32- and 56-Gbps for IB and 10- and 40- Gbps for HSE), these networks provide various other features too, such as hardware protocol offload, remote memory access capabilities, hardware multicast, Quality of service, rate control, multi-pathing, fault tolerance and path migration capabilities. Owing to such capabilities, these architectures are quickly being adopted by many high performance computing clusters as well as enterprise datacenter platforms. Products with varying levels of hardware support are also becoming available. At the same time, multi-core computing platforms with varying architectures are emerging. Thus, current and future network architectures provide new ways to design next generation cluster computing systems with multi-core architectures.

Agenda:

The agenda for the tutorial will be as follows: The presenter will provide introductory material and background to the audience. The topics of discussion will gradually advance from basic to intermediate to advanced. Following the introduction, the architectural features of IB and HSE will be described. Then, the current status and trends of cluster design with IB and HSE will be discussed. Several current hardware/software solutions will be described. Finally, sample performance numbers from various environments such as MPI, UPC, File Systems, Memcached and Cloud Computing (Hadoop, HDFS and HBase) will be shown.

Goals:

This tutorial will bring different network interconnects (with emphasis on IB and HSE) to the audience in a single coherent presentation, focus on their individual strengths and limitations and provide a comparative study between these standards. If the committee members have additional suggestions, the presenters will be happy to incorporate them.Based on these emerging trends and the associated challenges, the goals of this tutorial are as follows:

1. Making the attendees familiar with the IB and HSE network interconnects and the associated benefits.

2. Demonstrating how the OpenFabrics stack is trying to provide a convergence between these two standards.

3. Providing an overview of available IB and HSE hardware/software solutions.

4. Outlining case studies of designing next generation systems (HPC with MPI, PGAS/UPC and OpenSHMEM, File Systems and emerging enterprise datacenters and cloud computing environments with Memcached and Hadoop) while taking advantage of IB and HSE features and multi-core computing platforms.

In summary, the tutorial is aimed at providing a critical assessment of whether these technologies are ready for prime-time or not.

Targeted Audience and Expected Prerequisite Knowledge:

This tutorial is targeted for various categories of people working in the areas of high performance communication and I/O, storage, networking, middleware, virtualization, and applications related to clusters:

1. Newcomers to the field of high-speed networking who want to familiarize themselves with IB and HSE technologies.

2. Scientists, engineers, and researchers working on the design and development of next generation high-end systems including computing clusters, data centers, storage centers and cloud computing environments.

3. Developers of next generation networked computing middleware and applications.

4. Managers and administrators responsible for setting-up next generation clusters and facilities in their organizations/laboratories.

There is no fixed prerequisite for this tutorial. As long as the attendee has a general knowledge in high performance computing, cluster computing, networking, storage, and the related issues, he/she will be able to understand and appreciate it. The tutorial is designed a way that allows the attendee to get exposed to the topics in a smooth and progressive manner.

Content Level:

The content level will be as follows: 20% beginner, 50% intermediate, and 30% advanced.

Speaker information:                

Name: Dr. Dhabaleswar K. (DK) Panda

Address: Dept. of Computer Science and Engineering

The Ohio State University, Columbus, OH 43210–1277

Email: panda@cse.ohio-state.edu

Phone: (614)–292–5199
Fax: Fax: (614)–292–2911
Web: http://www.cse.ohio-state.edu/_panda

Biography:

Dhabaleswar K. (DK) Panda is a Professor of Computer Science at the Ohio State University. He obtained his Ph.D. in computer engineering from the University of Southern California. His research interests include parallel computer architecture, high performance computing, communication protocols, files systems, network-based computing, and Quality of Service. He has published over 300 papers in major journals and international conferences related to these research areas. Dr. Panda and his research group members have been doing extensive research on modern networking technologies including InfiniBand, HSE and RDMA over Converged Enhanced Ethernet (RoCE). His research group is currently collaborating with National Laboratories and leading InfiniBand and HSE companies on designing various subsystems of next generation high-end systems. The MVAPICH/MVAPICH2 (High Performance MPI over InfiniBand, iWARP and RoCE) open-source software packages, developed by his research group (http://mvapich.cse.ohio-state.edu), are currently being used by more than 1,910 organizations worldwide (in 67 countries). This software has enabled several InfiniBand clusters (including the 5th and 7th ranked ones) to get into the latest TOP500 ranking. More than 108,000 downloads of this software have taken place from the project’s website alone. This software package is also available with the Open Fabrics stack for network vendors (InfiniBand and iWARP), server vendors and Linux distributors. Dr. Panda’s research is supported by funding from US National Science Foundation, US Department of Energy, and several industry including Intel, Cisco, SUN, Mellanox, QLogic, NVIDIA and NetApp. He is an IEEE Fellow and a member of ACM.  

Tutorial title:  

The Current Challenges of Supercomputer Programming

  Sasa Stojanovic, Dragan Bojic, Marko Misic, Milo Tomasevic, Miroslav Bojovic, Veljko Milutinovic, Michael Flynn, and Oskar Mencer,

 University of Belgrade, Serbia,

Stanford University, Palo Alto, US,

Maxeler, London, UK.

Major readings:

Stojanovic, S. et al, "A Survey of Modern Reconfigurable Computing," Proceedings of the IEEE ICIT Conference, Athens, Greece, March 19-21, 2012.

Mencer, O. Flynn, M., Milutinovic, V. Rakocevic, G. Stenstrom, P., Trobec, R., Valero, M. "PetaFlops vs PetaData" Communications of the ACM, 2012.

Mencer, O., Flynn, M, "The Maxeler Exascale Computing," Maxeler, London, UK, April 2012.

This tutorial can last any time from three to six academic hours. Previous versions were delivered several times for industry, as in-house presentations, in US (Intel, SUN, ...) and EU (Philips, Siemens, ...), and for universities in US (Alabama, Purdue, ...) and EU (Barcelona, Ljubljana, ...).

Abstract:

First, the modern programming paradigms are discussed, based on the approaches established by the educational freeware of major US universities. Second, a set of classification criteria is discussed and a novel classification of existing approaches is presented. Third, each existing approach is analyzed using the same template: (a) A figure explaining the architecture used and (b) Discussion of major characteristics, pros and cons, and possible avenues for future improvements. Forth, some slides are dedicated to CUDA programming. Fifth, most of the slieds are dedicated to programming of dataflow supercomputers in general and the Maxeler dataflow supercomputer in particular. Sixth, architectural details and programming examples are given for Maxeler applications in GeoPhysics and banking.