Parallel Computing at DESY Zeuthen. Introduction to Parallel Computing at DESY Zeuthen and the new cluster machines

Similar documents
HPC Middle East. KFUPM HPC Workshop April Mohamed Mekias HPC Solutions Consultant. Agenda

ABySS Performance Benchmark and Profiling. May 2010

Advanced Topics in High Performance Scientific Computing [MA5327] Exercise 1

Tech Computer Center Documentation

Cluster Network Products

Cluster Clonetroop: HowTo 2014

Introduction to High Performance Computing at UEA. Chris Collins Head of Research and Specialist Computing ITCS

The German National Analysis Facility What it is and how to use it efficiently

Introduc)on to Hyades

Introduction to High Performance Computing at UEA. Chris Collins Head of Research and Specialist Computing ITCS

Before We Start. Sign in hpcxx account slips Windows Users: Download PuTTY. Google PuTTY First result Save putty.exe to Desktop

Datura The new HPC-Plant at Albert Einstein Institute

KISTI TACHYON2 SYSTEM Quick User Guide

Sami Saarinen Peter Towers. 11th ECMWF Workshop on the Use of HPC in Meteorology Slide 1

Batch system usage arm euthen F azo he Z J. B T

HPC Architectures. Types of resource currently in use

Grid Engine - A Batch System for DESY. Andreas Haupt, Peter Wegner DESY Zeuthen

Our new HPC-Cluster An overview

SPINOSO Vincenzo. Optimization of the job submission and data access in a LHC Tier2

Introduction to Parallel Programming. Martin Čuma Center for High Performance Computing University of Utah

ICON Performance Benchmark and Profiling. March 2012

The GPU-Cluster. Sandra Wienke Rechen- und Kommunikationszentrum (RZ) Fotos: Christian Iwainsky

Sun Lustre Storage System Simplifying and Accelerating Lustre Deployments

Introduction to Parallel Programming. Martin Čuma Center for High Performance Computing University of Utah

RWTH GPU-Cluster. Sandra Wienke March Rechen- und Kommunikationszentrum (RZ) Fotos: Christian Iwainsky

RHRK-Seminar. High Performance Computing with the Cluster Elwetritsch - II. Course instructor : Dr. Josef Schüle, RHRK

Genius Quick Start Guide

NICs (dedicated for iscsi): NIC Driver version

Feedback on BeeGFS. A Parallel File System for High Performance Computing

InfoBrief. Platform ROCKS Enterprise Edition Dell Cluster Software Offering. Key Points

BlueGene/L. Computer Science, University of Warwick. Source: IBM

User Training Cray XC40 IITM, Pune

AcuSolve Performance Benchmark and Profiling. October 2011

iwarp Learnings and Best Practices

DATARMOR: Comment s'y préparer? Tina Odaka

Site Report. Stephan Wiesand DESY -DV

The National Analysis DESY

X Grid Engine. Where X stands for Oracle Univa Open Son of more to come...?!?

Introduction to Parallel Programming. Martin Čuma Center for High Performance Computing University of Utah

Vienna Scientific Cluster: Problems and Solutions

Pedraforca: a First ARM + GPU Cluster for HPC

SuperMike-II Launch Workshop. System Overview and Allocations

Introduction to Parallel Programming. Martin Čuma Center for High Performance Computing University of Utah

NAMD Performance Benchmark and Profiling. January 2015

AIST Super Green Cloud

CRFS: A Lightweight User-Level Filesystem for Generic Checkpoint/Restart

How to run applications on Aziz supercomputer. Mohammad Rafi System Administrator Fujitsu Technology Solutions

Unified Runtime for PGAS and MPI over OFED

Running Applications on The Sheffield University HPC Clusters

Study. Dhabaleswar. K. Panda. The Ohio State University HPIDC '09

Linux HPC Software Stack

SNAP Performance Benchmark and Profiling. April 2014

HIGH PERFORMANCE COMPUTING FROM SUN

Wednesday, August 10, 11. The Texas Advanced Computing Center Michael B. Gonzales, Ph.D. Program Director, Computational Biology

Computing with the Moore Cluster

HYCOM Performance Benchmark and Profiling

HPCC New User Training

Improving Application Performance and Predictability using Multiple Virtual Lanes in Modern Multi-Core InfiniBand Clusters

Architecting High Performance Computing Systems for Fault Tolerance and Reliability

Emerging Technologies for HPC Storage

MIGRATING TO THE SHARED COMPUTING CLUSTER (SCC) SCV Staff Boston University Scientific Computing and Visualization

The RWTH Compute Cluster Environment

Introduction to HPC Using zcluster at GACRC

A Case for High Performance Computing with Virtual Machines

Solaris Engineered Systems

Reduces latency and buffer overhead. Messaging occurs at a speed close to the processors being directly connected. Less error detection

Application-Transparent Checkpoint/Restart for MPI Programs over InfiniBand

Maxwell: a 64-FPGA Supercomputer

HPC Resources at Lehigh. Steve Anthony March 22, 2012

Exploiting Task-Parallelism on GPU Clusters via OmpSs and rcuda Virtualization

My operating system is old but I don't care : I'm using NIX! B.Bzeznik BUX meeting, Vilnius 22/03/2016

10-Gigabit iwarp Ethernet: Comparative Performance Analysis with InfiniBand and Myrinet-10G

How to Use a Supercomputer - A Boot Camp

Performance comparison between a massive SMP machine and clusters

ACEnet for CS6702 Ross Dickson, Computational Research Consultant 29 Sep 2009

Brief review of the HEPIX 2011 spring Darmstadt, 2-6 May

Introduction to HPC Using zcluster at GACRC

To Infiniband or Not Infiniband, One Site s s Perspective. Steve Woods MCNC

MPI History. MPI versions MPI-2 MPICH2

Design and Evaluation of a 2048 Core Cluster System

DESCRIPTION GHz, 1.536TB shared memory RAM, and 20.48TB RAW internal storage teraflops About ScaleMP

The Role of InfiniBand Technologies in High Performance Computing. 1 Managed by UT-Battelle for the Department of Energy

The cluster system. Introduction 22th February Jan Saalbach Scientific Computing Group

PART-I (B) (TECHNICAL SPECIFICATIONS & COMPLIANCE SHEET) Supply and installation of High Performance Computing System

Headline in Arial Bold 30pt. Visualisation using the Grid Jeff Adie Principal Systems Engineer, SAPK July 2008

Übung zur Vorlesung Architektur paralleler Rechnersysteme

HPCC User Group Meeting

GROMACS (GPU) Performance Benchmark and Profiling. February 2016

CPMD Performance Benchmark and Profiling. February 2014

Advances of parallel computing. Kirill Bogachev May 2016

Research on the Implementation of MPI on Multicore Architectures

Making Supercomputing More Available and Accessible Windows HPC Server 2008 R2 Beta 2 Microsoft High Performance Computing April, 2010

LS-DYNA Performance Benchmark and Profiling. April 2015

MILC Performance Benchmark and Profiling. April 2013

LS-DYNA Best-Practices: Networking, MPI and Parallel File System Effect on LS-DYNA Performance

Niclas Andersson

Managing Multi-user Windows (Citrix) from a System Manager's View

Minnesota Supercomputing Institute Regents of the University of Minnesota. All rights reserved.

Voltaire. Fast I/O for XEN using RDMA Technologies. The Grid Interconnect Company. April 2005 Yaron Haviv, Voltaire, CTO

STAR-CCM+ Performance Benchmark and Profiling. July 2014

Transcription:

Parallel Computing at DESY Zeuthen. Introduction to Parallel Computing at DESY Zeuthen and the new cluster machines Götz Waschk Technical Seminar, Zeuthen April 27, 2010

> Introduction > Hardware Infiniband M1000e blade center Schematic view, blade center Schematic view, storage network > Software Operating system MPI versions Batch System Debugging support Lustre > User access Batch jobs Known Problems > Summary Götz Waschk (DESY) Parallel Computing at DESY Zeuthen April 27, 2010 2 / 19

Introduction Parallel Computing at DESY > apenext Special Purpose Computer > Local Batch Farm with slow 1G-Ethernet connections > New Pax Clusters with Infiniband Götz Waschk (DESY) Parallel Computing at DESY Zeuthen April 27, 2010 3 / 19

Hardware New cluster hardware > Hardware installed in 1/2010 > 8 Dell PowerEdge M1000e Blade Centers > M3601Q 32-Port 40G Infiniband Switches > 16 Dell PowerEdge M610 Blade servers each 2 quad-core Intel Xeon E5560 CPUs @ 2.8GHz QDR 40 GBit/s Infiniband 24 GB Main memory DDR3 (1.3GHz) 2 2.5 SAS drives, 146GB, RAID0 > Total peak performance: 12 TFLOPS Götz Waschk (DESY) Parallel Computing at DESY Zeuthen April 27, 2010 4 / 19

Infiniband networking Hardware Infiniband > MPI communication 40 GBit/s inside blade center Connected to internal QDR Infiniband switch > Storage network access to Lustre file system 20 GBit/s connection per blade center Connected to older Voltaire DDR Infiniband switch Götz Waschk (DESY) Parallel Computing at DESY Zeuthen April 27, 2010 5 / 19

Hardware M1000e blade center Götz Waschk (DESY) Parallel Computing at DESY Zeuthen April 27, 2010 6 / 19

Hardware Schematic view, blade center pax01 pax02 pax03 40Gbit/s pax0e pax0f M3601Q 32-Port 40G IB Switch 20Gbit/s storage connection to Voltaire switch Götz Waschk (DESY) Parallel Computing at DESY Zeuthen April 27, 2010 7 / 19

Hardware Schematic view, storage network pax1 pax2 pax3 pax4 pax5 Voltaire ISR2010 Infiniband Switch pax6 pax7 pax8 plus0 plus1 plus2 plus3 plus4 Götz Waschk (DESY) Parallel Computing at DESY Zeuthen April 27, 2010 8 / 19

Software Operating system Software > Standard SL5.4 64 bit > Same software as on desktop and farm > Several MPI versions OpenMPI Mvapich/Mvapich2 Intel MPI test installation Götz Waschk (DESY) Parallel Computing at DESY Zeuthen April 27, 2010 9 / 19

Software MPI versions Open MPI > Open Source implementation of the MPI-1 and MPI-2 standards > Versatile, supports many network types, batch systems > Dynamic loading of plug-ins > Comes with SL5.4 > Testing on your workgroup server/desktop possible > Automatic selection of the right network transport currently broken, use mpirun --mca btl "ˆ udapl" > Extra builds for Intel and PGI compilers use ini to select the right version Götz Waschk (DESY) Parallel Computing at DESY Zeuthen April 27, 2010 10 / 19

Software MPI versions Mvapich > Mvapich/Mvapich2 are Infiniband ports of MPICH/MPICH2 > Needs MPD (multi purpose daemon) running on all nodes > Not integrated with batch system > Binaries only run on machines with Infiniband > Comes with SL5.4 as well > Builds for gcc and Intel compilers Götz Waschk (DESY) Parallel Computing at DESY Zeuthen April 27, 2010 11 / 19

Software MPI versions Intel MPI > Based on MPICH2 > Needs commercial or evaluation license > Installed for testing > No batch integration > Supports both gcc and Intel compilers Götz Waschk (DESY) Parallel Computing at DESY Zeuthen April 27, 2010 12 / 19

Software Batch System Batch system > SUN Grid Engine 6.2u5 > Tight integration of OpenMPI OpenMPI s mpirun uses qsub to start MPI processes MPI processes are SGE tasks All MPI processes have AFS token All MPI processes run under SGE s control > Same SGE instance as used in the farm Götz Waschk (DESY) Parallel Computing at DESY Zeuthen April 27, 2010 13 / 19

Software Debugging support Debugging support > Currently, no parallel debugger installed Might be purchased if demanded possible choices: Intel Cluster Toolkit, Alinea DDT, Totalview > Intel debugger 11.0 is available > Valgrind with OpenMPI support is installed Götz Waschk (DESY) Parallel Computing at DESY Zeuthen April 27, 2010 14 / 19

Software Lustre Lustre file system > Open Source parallel file system > 1 Meta Data Server, 4 Object Storage Servers > Version 1.8.2 test installation > Advantages: Scalable parallel access High performance, > 500MB/s per file server > Disadvantages: Stability issues Complicated administration Unclear future since Oracle takeover > Used as scratch and staging file system, no backup! Götz Waschk (DESY) Parallel Computing at DESY Zeuthen April 27, 2010 15 / 19

User access User access to the cluster machines > Accessible by members of the nic, that and alpha groups > Other users like PITZ or photon are welcome > 2 blade centers as interactive machines: pax0 and pax1 > 6 blade centers in the batch farm Götz Waschk (DESY) Parallel Computing at DESY Zeuthen April 27, 2010 16 / 19

User access Batch jobs Batch job submission Most important parameters: > #$ -pe mpi-pax? 128 > #$ -R y > #$ -l h_vmem=3g Parallel jobs on the farm: > #$ -pe multicore-mpi 8 for just 8 cores > #$ -pe mpi 40 for larger jobs with low communication overhead Götz Waschk (DESY) Parallel Computing at DESY Zeuthen April 27, 2010 17 / 19

User access Known Problems Known Problems > Open MPI has a slow MPI_Sendrecv_replace on Infiniband 1 > Batch system submission error: no suitable queues reservations > Unsolved hardware problems on some nodes open Dell support issues > No SGE integration of Mvapich/Intel MPI 1 https://svn.open-mpi.org/trac/ompi/ticket/2153 Götz Waschk (DESY) Parallel Computing at DESY Zeuthen April 27, 2010 18 / 19

Summary Further reading > https://dvinfo.ifh.de/cluster/ > https://dvinfo.ifh.de/batch_system_usage/ > zn-cluster mailing list Götz Waschk (DESY) Parallel Computing at DESY Zeuthen April 27, 2010 19 / 19