CLUSTER

The platform ABiMS offers resources around a dedicated data center (Marine Genomic Data Center).
This data center is destined to data storage and security, as well as their analyses.
It represents :
  • 1600 processors handled by Sun Grid Engine middle-ware (SGE),
  • a secure data storage (RAID) of 800 To,
  • a record robot with two LTO5 players for the daily record of critical data.
This infrastructure has been financed by multiple sources :
  • Région Bretagne - BioGenOuest
  • Conseil Général - CG29
  • ANR projects 
  • Station Biologique de Roscoff: FR2424, UMR7139, UMR7144, UMR7150
  • Investissements d'Avenir projects: EMBRC-France, IDEALG, Océanomics, IFB
  • GRSIBI projects.
In 2013, 90% of the CPU resources involved collaborative projects with at less one partner from the "Station Biologique de Roscoff"; 10 % were used for fully external projects.

RESOURCES

This group structure allows to submit jobs on a specific computer type.
 
DIFFERENTS TYPES OF NODE
Group Node Type Function Processor RAM
@@bigcpu n40-n55 Dell blade M630 Multithreaded treatements

Intel

32 x 2.4 GHz

128 Go
  n60-n75 Dell blade M605

Reserved for some applications

AMD

8 x 2.4 GHz

32 Go
@@bignode n76-n79 Dell R815

Multithreaded treatements

Memory usage

AMD

48 x 2.2 GHz

256 Go
  n55-n59  

Multithreaded treatements

Memory usage

AMD

48 x 2.2 GHz

256 Go
@@bigcpu n80-n95 Dell C6220 Multithreaded treatements

Intel

32 x 2.2 GHz

128 Go
@@bigmem n99 Dell R910 Memory usage

Intel

40 x 2.0 GHz

1 To
 
A queue system managed by the SGE scheduler allows the job submission and monitoring.
The available queues are :
 
  Time out Available resources Purpose
short.q 12 hours High priority - 50 % of all CPU (- n99) Regular jobs
long.q 10 days Standard priority - 50 % of all CPU (- n99) Long jobs
infinite.q no limit Low priority - 25 % of all CPU (- n99) Really really long jobs
bigmem.q     For treatements requiring a lot of RAM
qlogin.q 2 days   For connections in qlogin mode
clc.q   n76 For CLC use (please contact support.abims@sb-roscoff.fr).

A dedicated queue, named iprscan.q, is for InterProScan applications. It is automatically used by the application without having to use the qsub submission command.

 

MAXIMUM NUMBER OF SLOTS AVAILABLE PER QUEUE FOR EACH GROUP

  @@bignode @@bigcpu @@bigmem
short.q 24 16 0
long.q 24 16 0
infinite.q 12 8 0
bigmem.q 0 0 40

In case of competition during treatements in parallel (for example : jobs on three queues), treatements on the short.q queue have priority, followed by those running on long.q, and then by those running on infinite.q.