Special Feature Writeup of ISC09

A good collection of Articles and Commentaries dedicated to ISC 09 Conference
HPCwire's special section dedicated to ISC09

Grand unification of supercomputing and cloud computing?

There were a posting by Michael Feldman, editor of HPCwire on the title High Performance Cloud Computing Still an Oxymoron.

A good article where experts provide differing view on the feasibility of Cloud Computing on HPC infrastructure.

The concept of Processor Equivalent

Learned a new concept of Process Equivalent or PE. It is a measure of the actual impact of a set of requested resources by a job on the total resources system wide.

Taken from MAUI Admin Guide by Cluster Resources

PE =MAX (ProcsRequestedByJob / TotalConfiguredProcs,
MemoryRequestedByJob / TotalConfiguredMemory,
DiskRequestedByJob / TotalConfiguredDisk,
SwapRequestedByJob / TotalConfiguredSwap) * TotalConfiguredProcs

Assume a homogeneous 100 node system with 4 processors and 1 GB of memory per node. A job is submitted requesting 2 processors and 768 MB of memory. The PE for this job would be calculated as:

PE = MAX(2/(100*4), 768/(100*1024)) * (100*4) = 3.

This result makes sense since the job would be consuming 3/4 of the memory on a 4 processor node. The calculation works equally well on homogeneous or heterogeneous systems, uniprocessor or large way SMP systems.

Installing Intel Compiler on a compute node

The Compute Nodes of a HPC are often on a private network and may not have access to the internet and they are often very lean. How do we compile the Intel Compiler on the compute node?

After untaring the Intel Compiler, and running install.sh and registering the serial number, you will notice that there will be a missing essential prequiste. How to know what is the missing important prequiste?

Go to the intel installation log file /tmp/intel.pset.root........, you can see in the log file "missing g++". To install g++, do the following:

# yum install gcc-c++

Problem solved!

Multi-core Strategies: MPI and OpenMP

Multi-core Strategies: MPI and OpenMP is an interesting article on MPI and OpenMP from HPCommunity.org

pbs node manipulation Revision

It has been a long time since I used torque. I though I pen down some common commands I will used

• pbsnodes -l (List node names and their state. If no state is specified, only nodes in the DOWN, OFFLINE, or UNKNOWN states are listed. Specifying a state string acts as an output filter. Valid state strings are "active", "all", "busy", "down", "free", "offline", "unknown", and "up". )
  
• pbsnodes -o (Add the OFFLINE state. This is different from being marked DOWN. OFFLINE prevents new jobs from running on the specified nodes. This gives the administrator a tool to hold a node out of service without changing anything else. The OFFLINE state will never be set or cleared automatically by pbs_server; it is purely for the manager or operator.)
  
• pbsnodes -c (Clear OFFLINE from listed nodes)
  
• pbsnodes -a (All attributes of a node or all nodes are listed. This is the default if no flag is given)
  

Optimizing OpenFOAM Productivity in Clusters - Article from Desktop Engineering

This is an interesting article from Desktop Engineering "Optimizing OpenFOAM Productivity in Clusters". This article explore the performance gained when the HPC system is configured properly, there is a lot of productivity and gain in performance.

libibverbs: Warning: RLIMIT_MEMLOCK is 32768 bytes. This will severely limit memory registrations.

We are using OpenFarics for our HPC. According to the FAQ, "With OpenFabrics (and therefore the openib BTL component), you need to set the available locked memory to a large number (or better yet, unlimited) -- the defaults with most Linux installations are usually too low for most HPC applications that utilize OpenFabrics"

libibverbs: Warning: RLIMIT_MEMLOCK is 32768 bytes.
This will severely limit memory registrations.

For the solution, look here at http://www.open-mpi.org/faq/?category=openfabrics#ib-locked-pages
Or you can also look at my blog entry Encountering Segmentation Fault, Bus Error or No output

Installing MPI4py on CentOS

MPI for Python (mpi4py) provides bindings of the Message Passing Interface (MPI) standard for the Python programming language, allowing any Python program to exploit multiple processors.



Step 1: Install Python and dependencies
# yum install python python-devel python-lib

Step 2: Build OpenMPI with Intel Compiler
See Blog Entry Building OpenMPI with Intel Compiler (Ver 2)

Step 3: Get and install MPI4py
# wget http://mpi4py.googlecode.com/files/mpi4py-X.X.X.tar.gz

Step 4: Build the MPI4py

# cd /usr/local

# tar -zxf mpi4py-X.X.X.tar.gz

# cd mpi4py-X.X.X

Step 5: Setup of MPI
# python setup.py build
# python setup.py install

Step 6: Testing
Issuing at the command line:
$ mpiexec -n 5 python demo/helloworld.py
(will launch a five-process run of the Python interpreter and run the test scripts)

Important Notes
For more information on the installation of MPI4py, see the official MPI4py User Guide

Compiling with MPI Parallel Library

This is a good site for MPICH Parallel Library Compilation from Duke University

Advancing the Power of Visualization.

This is an interview by HPwire with Steve Briggs, HPCD’s SVA product marketing manager on Visualisation from HP point of view. Interesting information

Advancing the Power of Visualization –Coming Soon to Linux Clusters: 100 Million Pixels and More

MPI Tutorials Information

Free Online MPI Tutorials and documentation

1. Intro to MPI Course Overview
2. Parallel Programming with MPI
   

Moving HPC Applications to Cloud - The Practitioner Prospective

This is a very good summarise presentation by Victoria Livschitz, CEO of Grid Dynamics on some of the issues and challenges we will face when we unify Cloud and HPC into HPC-Cloud.

Read this: Moving HPC Applications to Cloud - The Practitioner Prospective

Intel® Optimized LINPACK Benchmark for Linux OS

This blog entry is taken from Intel(R) Math Kernel Library for the LINUX* OS User's Guide. This document comes when you download and install the Math Kernel Library.

To download, see Intel® Math Kernel Library – LINPACK Download

Intel® Optimized LINPACK Benchmark is a generalization of the LINPACK 1000 benchmark.
It solves a dense (real*8) system of linear equations (Ax=b), measures the amount of
time it takes to factor and solve the system, converts that time into a performance rate, and tests the results for accuracy. The generalization is in the number of equations (N) it can solve, which is not limited to 1000. It uses partial pivoting to assure the accuracy of the results.

Intel is providing optimized versions of the LINPACK benchmarks to make it easier than using HPL for you to obtain high LINPACK benchmark results on your systems based on genuine Intel® processors. Use this package to benchmark your SMP machine.



1. Running the Software
To Run pre-determined sample problem sizes on a give system
# ./runme_xeon32 OR
#./runme_xeon64
To run problem for other problem sizes, you can use and amend liniput_xeon32, and liniputxeon64. However each input file requires the following amount of memory:

• lininput_xeon32 2GB
• lininput_xeon64 16GB

2. Known Limitation

• Intel LINPACK Benchmark is threaded to effectively use multiple processors. In multi-processor systems, best performance will be obtained with Hyper-Threading Technology turned off.
• If an incomplete data input file is given, binaries may either hang or fault

IBM Deep Computing for High Performance Computing (HPC) Site

Good Resource from IBM Deep Computing

1. IBM Deep Computing for High Performance Computing (HPC)
2. Good Bi-Monthly Article on IBM HPC "IBM HPC Insights"
3. IBM-HPC FaceBook Links http://www.facebook.com/pages/IBM-HPC/283405154323

Error: Missing Dependency: librrd.so.2()(64bit) is needed by package ganglia-gmetad (epel)

This is the continuation of Blog Entry Error: Missing Dependency: rrdtool = 1.2.27-3.el5 is needed by package rrdtool-perl

1. When I'm installing Ganglia on my CentOS, I've did the following:
   # yum install rrdtool ganglia ganglia-gmetad ganglia-gmond ganglia-web httpd php
   
   However, instead I got the followings "Error: Missing Dependency: rrdtool = 1.2.27-3.el5 is needed by package rrdtool-perl"
   
2. To resolve it, similar to the earlier Blog Entry, I install rrdtool first. But when I use the command
   # yum install ganglia ganglia-gmetad ganglia-gmond ganglia-web httpd php
   
   There is still error "Error: Missing Dependency: librrd.so.2()(64bit) is needed by package ganglia-gmetad (epel)...."
   
3. To resolve the issue, you have to manually get rid of the latest rrdtool (x86-64) installed by EPEL and install the rrdtool-1.2.27-3.el5.x86_64
   # yum remove rrdtool-1.3.8-2.el5.rf.x86_64
   # yum install rrdtool-1.2.27-3.el5.x86_64
   
   
4. Install the Ganglia and its related tools
   # yum install ganglia ganglia-gmetad ganglia-gmond ganglia-web httpd php

Implementation of HPC as a Service - the Shanghai Supercomputer Center

The Shanghai Supercomputer Center (SSC) was founded in December 2000. It is the first high performance computing platform open to the whole public in China and currently is the country’s leading supercomputer centre.......

Read more on HPC as a Service from Scientific Computing World

Sun HPC ClusterTools 8.2.1

Sun HPC ClusterTools 8.2.1 software is an integrated toolkit that allows developers to create and tune Message-passing Interface (MPI) applications that run on high performance clusters and SMPs. Sun HPC ClusterTools 8.2.1 is based on OpenMPI v1.3.4

OpenMPI 1.4 is ready for download

OpenMPI v1.4 is ready for download. According to the OpenMPI v1.4 README file, the different between v1.4 and the previous version 1.3.4 is the fix for Libtool 2.2.6b's issues.

Undergraduate Petascale Education Program

Petascale Education Program seeks to promote understanding and petascale computing among undergraduate and faculty. On the site, you can fund useful materials, undergraduate petascale modules etc....