Vous êtes ici: UNIL > Centre informatique > Services au personnel > Calcul > Infrastructures

Infrastructures

Argos | Hippocrate | Grid computing | CADMOS
 

Types of scientific computing resources available at the Ci: 

The new features are higlighted in blue 

Argos

Argos is an Intel Xeon based 8 cores interactive compute server offering GPGPU capabilities:

 

Job category --> Any suitable interactive job

Access  --> via ssh to argos1.unil.ch

OS         --> Redhat ELS 6.1
CPUs      --> 1 x dual Xeon 5570 2.97 GHz node (8 cores), 24 GB RAM
GPGPU   --> 1 x Nvidia Tesla c1060 (220 CUDA cores)
                   1 x Nvidia Tesla c2070 (448 CUDA cores)
Network  --> 10 Gb/s Ethernet

 

TOP ^

Hippocrate

Hippocrate is a small 96 cores Intel Xeon based HPC cluster:

 

Job category --> - Any suitable parallel job or any sequential long (< 96h) running job
                         - A maximum of 1000 jobs per user in the queuing system


Access               --> via ssh to argos1.unil.ch

OS                    --> RedHat ELS 6.1

 

Nodes --> 8 x dual Xeon 5670 2.97 GHz (12 cores), 48 GB RAM for batch processing
                1 x dual Xeon 5670 2.97 GHz (12 cores), 96 GB RAM for interactive jobs
               3 x Nvidia Tesla m2090 (512 CUDA cores) on 2 batch and 1 interactive nodes

 

 


Interconnect       --> Infiniband 40 Gb/s
Network             --> 10 Gb/s Ethernet

Parallel env.       --> - SMP (on 12 cores)
                               - PVM
                               - MPI : OpenMPI, MVAPICH

 

Batch schedduler      --> Grid Engine v6.2u5

Submission queues   --> - all.q (the default one)

* max 72 monothreaded compute slots
* max 3.9 GB of memory per slot (RAM + VRAM)
* max 96 hours of Wall Clock Time per slot 

                                      - short.q ( add  " #$ -q short.q "  in your submission script) 

* max 96 monothreaded compute slots
* max 3.9 GB of memory per slot (RAM + VRAM)
* max 24 hours of Wall Clock Time per slot


                                     - int.q  is an interactive submission queue

* max 12 interactive compute [threads|slots]
* max 7.9 GB of RAM + VRAM per [threads|slots]
* max 24 hours of Wall Clock Time per [threads|slots]

* type "qrsh -q int.q -pe smp 1" on argos1 to open a
    remote interactive shell with 1 [thread|slot]
* in the "-pe smp n" argument, n MUST match the the number
   of threads your application will use!
* Type Ctl-D to release the slots

 

TOP ^

Grid computing

The users of UNIL have also access to two grid-based infrastructures:

GridUNIL        --> a Condor based campus grid
                     --> ~ 300 distributed cores mainly on OSX and some on Linux or Windows platform
                     --> a grid distributed version of the R statistic environment
Job category   --> any suitable short (< 12h) embarrassingly parallel jobs


SMSCG           --> the Swiss Multisciences Computing Grid
                      --> ~ 4000 cores mainly on Intel/Linux platform
Job category    --> depends on the type of the targeted grid enabled resources

 

Grid submission clients:

Access                   --> via argos1.unil.ch

Condor-G               --> GridUNIL, SGE HPC cluster, SMSCG and many other grid Infrastructures
ARC client              --> SMSCG and many other ARC based grid Infrastructures
Globus client          --> GridUNIL and many other Globus based grid Infrastructures
Ganga client (GUI)  --> GridUNIL, SGE HPC cluster, SMSCG and many other grid Infrastructures 

TOP ^

CADMOS

CADMOS is the Center for Advanced Modeling Science. It is a joint initiative between UNIL EPFL and UNIGE. They offer an access to massively parallel computing resources currently based on an IBM BlueGene/P

TOP ^


Recherche:
 dans ce site:
   
   
   
 Rechercher
Annuaires      Site map

Amphimax  -  CH-1015 Lausanne  -  Suisse  -  Tél. +41 21 692 22 11  -  Fax  +41 21 692 22 05