Computing Service Legacy

From Nano Group Budapest
Jump to: navigation, search

THIS PAGE CONTAINS NOT SUPPORTED LEGACY INFORMATION

Introduction

Security Tips

First, please follow our security advises:

  • Use a Unix-based/like operating system. We support and recommend OS X, SuSE and Ubuntu Linux.
  • Use automatic OS updates frequently.
  • Use a strict firewall and apply strong passwords. There is a console-based password manager for random passwords in Shpak.
  • Keep your ssh private keys in safe and use string key passwords as well.
  • Setup and use GPG encrypted mails and chat.
  • Use strong encryption for your private date.
  • Use strong encryption for cloud-based drives as well. Shpak has support for directory encryption by the Fuse encfs module. If you need interoperability try Wuala.

A good general guide on privacy can be found at eff.org. OS X users should follow Apple's security manual. Linux users can find security notices here for SuSE and for Ubuntu. Please also subscribe to the corresponding security mailing lists.

For a more detailed security advises please see our Notes on Security page.


SZFKI System

Skynet

This is our main SGI cluster installed in May 2011. A late descendant of the venerable TPA batch systems and the CEDRUS environment.

You can find a large collection of documents and manuals on using SGI systems in SGI's Techpubs Library. We recommend the application tuning and the SGI MPT part.

Specification

  • CPU: Intel Xeon E5620 @ 2.40GHz SMT off
  • Cores per CPU: 4
  • Available Memory per Node: 34GB
  • Architecture: x86_64 / intel64 / em64t little-endian
  • Total number of nodes: 26
  • Allocation Unit: 1 node (8 cores)
  • Scheduler: Slurm
  • Purpose: Materials Science and Nanotechnology

Projects

Project ID Priority Description Participiants Collaborators
diamond high Hugo Pinto, Tamás Simon
sic normal Viktor Ivády, Krisztián Szász, Bálint Somogyi, Tamás Hornos
diavib low Márton Vörös, Tamás Demjén
solar low Márton Vörös

Shell Environment

Setup the Shell

You can fine-tune the shell by defining aliases and functions for frequently used commands. System default setup can be sourced in $HOME/.profile by:

 source /data1/ngb/site/bin/skynet

Install Shpak

Shpak is our ultimate shell toolkit. Regular job submission is supported only by Shpak. It promotes a unified and clean interface for various queue systems and simulation programs. To install use git in your $HOME:

 git clone git://github.com/hornos/shpak.git

and set the path in your $HOME/.profile:

 PATH=$PATH:$HOME/shpak/bin

Install Shpak (full)

Alternatively, you can use our 3-component package group for ab initio people:

 git clone git://github.com/hornos/shpak.git
 git clone git://github.com/hornos/qcpak.git
 git clone git://github.com/hornos/pypak.git

in this case you have to source qcpak only in your $HOME/.profile:

 source $HOME/qcpak/qcpakrc &> /dev/null
 PS1_HOST=SKYNET
 ps1

The last two command sets the prompt with a label PS1_HOST. You can switch between the simple and the advanced version by the command

 ps1

Storing Files

Use compression in every case to save storage space! You can use the parallel compressor programs pigz or pbzip2 as a replacement for gzip or bzip2, respectively. You can configure Shpak to use these programs by creating a user defined config file in $HOME/shpak/lib/z/usr.cfg with the following content:

 sp_b_z="pigz -9 -f"
 sp_b_uz="pigz -f -d"

Transfering Files

Regular users can use Shpak's transfer capabilities between xfer directories. First, create the remote xfer directory on each machine by:

 sshcmd -m <MID> -x "mkdir xfer"

Copy to remote machine, where <SOURCE> is the local path to be copied.:

 sshpush -m <MID> -s <SOURCE>

Copy from the remote machine:

 sshpull -m <MID> -s <SOURCE>

Transfer mode can be set by -t <MODE>, where <MODE> is

  1. Regular SCP transfer
  2. Tar files and use SCP for the transfer
  3. Rsync over SSH

To sync large amount of data use transfer mode 3.

Users of the fuse group can mount remote systems by sshfs. At first, you have to prepare the local sshfs directory:

 mkdir /local/home/${USER}/sshfs
 ln -s /local/home/${USER}/sshfs ${HOME}/sshfs

Mounting a remote machine:

 sshmnt -m <MID>

Unmounting:

 sshumnt -m <MID>

After a not clean unmount a lock will remain preventing further mounts. Use -f option to clear the lock.

Storage

Enable ESZR-related modules in your $HOME/.modenv:

 module use /site/eszr/mod/common
 module load eszr/site
 module load eszr/sys/skynet

Check ESZR environment by:

 eszr_info

Your storage data directory is $ESZR_DATA . Storage is accessible only form the login node:

 cd $ESZR_DATA

Backup

By omitting xfer directory in a machines's MID file (sp_p_scp_remote variable) you can transfer files or directories directly to or from your remote home. If you want to synchronize from a remote machine to your current directory on your local machine:

 sshsync -m <MID> -s "<SOURCES>"

where <SOURCES> is a space-separated list of files or directories in your remote home. If you want to synchronize local files or directories from your current directory to a remote machine:

 sshsync -p -m <MID> -s "<SOURCES>"

sshsync uses Rsync over SSH (transfer mode 3) best for large backups. You do not have to specify MID and SOURCES every time if you create a .sshsync file with following content:

 _m=<MID>
 _s="<SOURCES>"

If you want to do a full remote home backup then omitt <REMOTE USER>/xfer in the MID file and

 sshsync -m <MID> -s <REMOTE USER>

Large backups takes a long time at first. you can use Shapk's screen wrapper to open a new screen shell session and let the backup run uninterrupted. To open or reopen a new screen:

 scrsel

Leave the session by Ctrl+A Ctrl+D. To enter type the command again and select a session.

Module System

To activate the module system either type modenv or create the $HOME/.modenv file with your desired module commands, eg:

 module use /data1/ngb/modulefiles
 module load sys/skynet
 module load env/sgi

Basic Module Commands

Short alias commands are also available if you use Shpak. The module system resolves dependencies or conflicts automatically thus explicit unload is not necessary.

Command Shpak alias Descritpion
module avail mla Show available modules
module list mls List loaded modules
module display mdp About the module
module load/unload <MODULE> mld/mlu <MODULE> Load / unload a module

MPI Subsystem

The recommended MPI subsystem is SGI MPT. You can check the MPI subsystem either by module list or by which mpirun. MPI environments are mutually exclusive and environment modules unload each other.

SGI MPT

Versions with mpt suffix are compiled with SGI MPT. You can load the environment and the corresponding program version by:

 module load env/sgi
 module load <PROGRAM>/<VERSION>.mpt

If there is a tuned configuration for the <PROGRAM> please load the tuner configuration as well:

 module load <PROGRAM>/tuner.mpt

Jobfile

Enable SGI MPT in the jobfile:

 MPIRUN="mpt"

and also set the following for binding:

 MPT_BIND="dplace -s 1"

Scheduler

The job scheduler is Slurm. In Slurm each user is assigned with one or more account which you have to set in the queue file.

General information about the partitions:

 sinfo -l
Partition Allowed Groups Purpose
service pexpress Small and short jobs / Development
batch pbatch General production

General information on jobs:

 squeue -l

List your jobs:

 myqstat

Pending job priorities:

 sprio -l

Slurm accounts and priorities:

 sshare -l

Job Submission

Shpak has a wrapper for job submission as well as running various programs. You need 3 components:

  1. Queue file: permanent parameters for the scheduler and the queue
  2. Job file: job specific resource parameters
  3. Guide file: program specific parameters for Shpak's runprg command

Queue file

Queue files are in $HOME/shpak/que/. You have to create a queue file for each scheduler/queue pair only once.

Key Value Descritpion
SCHED slurm / sge / pbs Scheduler type
QUEUE_MAIL_TO your@email Your email address
QUEUE_MAIL ALL / abe Set scheduler email notifications
QUEUE_PROJECT string Slurm account (project id)
QUEUE_CONST list Space-separated list of constraints
QUEUE_PART string Slurm partition
QUEUE_QUEUE string Scheduler queue
QUEUE_QOS string QOS type
QUEUE_SETUPS string Space-separated list of queue setup scripts
QUEUE_ULIMIT commands Ulimit commands

Example queue file ($HOME/shpak/que/example):

 SCHED="slurm"
 QUEUE_MAIL_TO="<YOUR@EMAIL>"
 QUEUE_MAIL="ALL"
 QUEUE_PROJECT="<PROJECT ID>"
 QUEUE_CONST="ib"
 QUEUE_SETUPS="/data1/ngb/site/bin/jobsetup"
 QUEUE_ULIMIT="ulimit -s unlimited; ulimit -l unlimited"

Job file

You need a job file to submit a job by the jobsub command. The job file refers to the above discussed queue file.

Valid job KEY description
Key Value Descritpion
NAME string Job name
TIME hh:mm:ss Wallclock time limit
MEMORY integer Memory per core limit in MB
NODES integer Number of nodes
SCKTS integer Number of CPU sockets
CORES integer Number of cores per socket
QUEUE string Name of the queue file
MPIRUN mpt / ompi / impi MPI subsystem type
MPT_BIND omplace / dplace SGI MPT CPU bind command (dplace or omplace)
OMPI_BIND -by* -bind-to-* Open MPI CPU bind options
SETUPS list Space-separated list of custom setup scripts
COMMAND command Command or script to submit

Example job file:

 NAME=test
 TIME=00:30:00
 MEMORY=2000
 NODES=2
 SCKTS=2
 CORES=4
 QUEUE=example
 MPIRUN="mpt"
 MPT_BIND="dplace -s 1"
 COMMAND="runprg -p vasp -g vasp.guide"


NIIF Systems

Debrecen

This a an SGI Altix ICE8400EX system maintained by NIIF.

Specification

  • CPU: Intel Xeon X5680 @ 3.33GHz SMT on
  • Cores per CPU: 6
  • Available Memory per Node: 48GB
  • Architecture: x86_64 / intel64 / em64t little-endian
  • Allocation Unit: 1 core
  • Scheduler: SGE
  • Purpose: General

Shell Environment

Setup the Shell

We have a prepared and precompiled environment which you can sync from Skynet. Follow these steps:

1. Synchronize the environment

 cd /share/niif/debrecen.legacy
 sshsync -p -m <DEBRCEN MID>

2. Login to the Debrecen machine and enable git

 PATH=$PATH:$HOME/local/usr/bin

3. Install paks

 git clone git://github.com/hornos/shpak.git
 git clone git://github.com/hornos/qcpak.git
 git clone git://github.com/hornos/pypak.git

4. Relogin

Module System

To activate the module system edit your $HOME/.profile:

 module purge
 module use ${HOME}/local/modulefiles
 module load sys/debrecen
 module load env/sgi
 # other modules ...

Scheduler

The job scheduler is SGE.

General information about the queues:

 qstat -f -g c

General information on jobs:

 qstat

Job submission

If you use Shpak job submission is almost the same as on Skynet. Differences are the following.

Queue file

An example queue file:

 SCHED="sge"
 QUEUE_MAIL_TO=<YOUR@EMAIL>
 QUEUE_MAIL="abe"
 QUEUE_QUEUE="debrecen.q"
 QUEUE_PE="mpi"
 QUEUE_EXCLUSIVE="yes"
 QUEUE_SHELL="/bin/bash"
 QUEUE_OPTS="-cwd -j y -V"
 QUEUE_SETUPS="${HOME}/local/bin/jobsetup"

Job file

Uncomment MEMORY requirement:

 # MEMORY=...

This machine has 6 cores per socket:

 CORES=6


Szeged

This a HP CP4000BL blade cluster system maintained by NIIF.

Specification

  • CPU: AMD Opteron 6174 @ 2.20 Ghz
  • Cores per CPU: 12
  • Available Memory per Node: 124GB
  • Architecture: x86_64 / intel64 / em64t little-endian
  • Total number of nodes: 48
  • Allocation Unit: 1 core
  • Scheduler: SGE
  • Purpose: General

Shell Environment

Setup the Shell

We have a prepared and precompiled environment which you can sync from Skynet. Follow these steps:

1. Synchronize the environment

 cd /share/niif/szeged.legacy
 sshsync -p -m <SZEGED MID>

2. Login to the Szeged machine and enable git

 PATH=$PATH:$HOME/local/bin

3. Install paks

 git clone git://github.com/hornos/shpak.git
 git clone git://github.com/hornos/qcpak.git
 git clone git://github.com/hornos/pypak.git

4. Relogin

Module System

To activate the module system edit your $HOME/.bash_profile:

 module purge
 module use ${HOME}/local/modulefiles
 module load sys/szeged
 module load env/intel
 # other modules ...

Scheduler

The job scheduler is SGE.

General information about the queues:

 qstat -f -g c

General information on jobs:

 qstat

Job submission

If you use Shpak job submission is almost the same as on Skynet. Differences are the following.

Queue file

An example queue file:

 SCHED="sge"
 QUEUE_MAIL_TO=<YOUR@EMAIL>
 QUEUE_MAIL="abe"
 QUEUE_QUEUE="szeged.q"
 QUEUE_PE="mpi"
 QUEUE_EXCLUSIVE="yes"
 QUEUE_SHELL="/bin/bash"
 QUEUE_OPTS="-cwd -j y -V"
 QUEUE_SETUPS="${HOME}/local/bin/jobsetup"

Job file

Uncomment MEMORY requirement:

 # MEMORY=...

This machine has 12 cores per socket and 4 CPUs per node:

 NODES=1
 SCKTS=4
 CORES=12


Pécs

This a an SGI Altix UV system maintained by NIIF.

Specification

  • CPU: Intel Xeon X7542 @ 2.67GHz SMT off
  • Cores per CPU: 6
  • Available Memory: 6TB
  • Architecture: x86_64 / intel64 / em64t little-endian
  • Total number of CPUs: 1151
  • Allocation Unit: 1 core
  • Scheduler: SGE
  • Purpose: General

Shell Environment

Setup the Shell

NSC Systems

Neolith

This a Linux cluster maintained by NSC.

Specification

  • CPU: Intel Xeon E5345 @ 2.33 GHz SMT off
  • Cores per CPU: 4
  • Available Memory per Node: 16/32GB
  • Architecture: x86_64 / intel64 / em64t little-endian
  • Total number of nodes: 805
  • Allocation Unit: 1 core
  • Scheduler: Slurm
  • Purpose: General
  • User Guide

Shell Environment

Setup the Shell

We have a prepared and precompiled environment which you can sync from Skynet. Follow these steps:

1. Synchronize the environment

 cd /share/nsc/neolith.nsc.liu.se
 sshsync -p -m <NEOLITH MID>

2. Login to the Neolith machine and enable git

 PATH=$PATH:$HOME/local/usr/bin

3. Install paks

 git clone git://github.com/hornos/shpak.git
 git clone git://github.com/hornos/qcpak.git
 git clone git://github.com/hornos/pypak.git

4. Relogin

Module System

NSC uses a rather old module system. Please use the provided .bash_profile or refer to the User Guide if you have special needs.

Scheduler

The job scheduler is Slurm. For basic commands refer to the User Guide.

Job submission

If you use Shpak job submission is almost the same as on Skynet. Differences are the following.

Queue file

An example queue file:

 SCHED="slurm"
 QUEUE_MAIL_TO=<YOUR@EMAIL>
 QUEUE_MAIL="ALL"
 QUEUE_PROJECT="<PROJECTID>"
 QUEUE_SETUPS="${HOME}/local/bin/jobsetup"


PRACE Systems

PRACE systems use a certificate based SSH from the Globus Toolkit. You can load to it on Skynet by:

 mld globus/5.0.4

Static binaries for your local machine can be downloaded from LRZ. Shpak commands start with gssh employs the Globus method. Currently, FUSE sshfs mount is not supported with certificates and you have to use push, pull or sync (see Transfering Files).

Huygens

This is an IBM Power6 cluster maintaind by SARA.

Setup the Shell

We have a prepared and precompiled environment which you can sync from Skynet. Follow these steps:

1. Synchronize the environment

 cd /share/prace/huygens.sara.nl
 gsshsync -p -m <HUYGENS MID>

2. Install paks

 git clone git://github.com/hornos/shpak.git
 git clone git://github.com/hornos/qcpak.git
 git clone git://github.com/hornos/pypak.git

3. Relogin

PRACE Environment

The prace module defines global variables which you can use on all sites. Check the variables by:

 set | grep PRACE

For PRACE-related services (eg.: internal gridftp) there is a utility script. Check the help by:

 prace_service -h