
Computing infrastructure
Two computing clusters remain at the staff’s disposal.
HARDWARE SPECIFICATION
zero.ift.uni.wroc.pl
Number of nodes | 7 |
Processor | Quad-Core AMD Opteron(tm) Processor 2384 |
Number of sockets | 2 |
Number of cores | 4 |
Number of threads | 1 |
Clock rate | 2.7 GHz |
Ram | 16 GB |
Home catalogues | /home | 872G | NFS resource |
rei.ift.uni.wroc.pl
Number of nodes | 2 |
Processor | Intel(R) Xeon(R) CPU E5-2620 v3 |
Number of sockets | 2 |
Number of cores | 6 |
Number of threads | 2 |
Clock rate | 2.4 GHz |
Ram | 64 GB |
Home catalogues | /home | 985G | NFS resource |
Scratch | /scratch | 30G | Local storage |
Queuing system
To use the computing nodes it is necessary to prepare a PBS program script and put it in the task queue (eg. qsub ./skryptPBS.sh) or use the possibility of interactive access to the shell on the computing node. Below we present basic information regarding the PBS system and an exemplary script. A detailed description of command operations is available upon inputing the command man COMMAND on the access server.
JOB MANAGEMENT
- qsub: order task
- qdel: delete task
- qhold: hold task
- qrls: resume held tasks
- qrerun: re-run task
- qmove: move task to another queue
TASK MONITORING
- qstat: show status of batch tasks
- qselect: show a specific set of tasks
- showq: shows current state of the claster
TASK ATTRIBUTES
Attributes of the PBS system for tasks can be set in two ways:
- via command line arguments to qsub
- through PBS directives in scripts passed to qsub (RECOMMENDED!)
Attribute | Value | Description |
-l | number of resources needed (np.nodes=1,walltime=00:30:00) | Defines needed resources and limits the number of resources that can be used. |
-N | task name | Defines the task’s name |
-o | file path | Defines the path to which the task’s standalone output (STDOUT) will be routed |
-e | file path | Defines the path to which the error output (STDERR) of the task will be routed |
-p | integer in the interval [-1024,1023] | Defines the task’s priority. Higher value indicates higher priority. |
-q | queue name | Defines the queue in which the task needs to be put |
Type of resource | Value | Description |
nodes | number of nodes | Number of nodes to use, it is also possible to specify the number of processes per node, eg. nodes=1:ppn=8 |
walltime | gg:mm:ss | Specification of estimated time per task |
cput | gg:mm:ss | Maximum processor time per task |
mem | integer value with appropriate suffix b,kb,mb,gb | Maximum RAM per task |
ncpus | positive integer | Declaration of the number of CPUs required |
INTERACTIVE MODE
To use interactive node access, call the command qsub -I -q shell -l nodes=1
Interactive mode is only available in the shell queue, it allows accessing the node directly via ssh – which is especially useful when configuring new software or debugging programs.
EXEMPLARY SCRIPT
To put a task in the queue using script, call the command qsub ./skrypt.sh
The simplest script to use in our cluster (it can be also found in our home catalogues ~/skrypt.sh)
#!/bin/bash
#### We define the task name
#PBS -N helloWorld
#### We will redirect output to the folder /scratch
#### Accelerate performance in operations I/O
#PBS -o /scratch/$USERNAME/helloWorld.stdout
#PBS -e /scratch/$USERNAME/helloWorld.stderr
#### Place in the appropriate queue
#PBS -q oneDay
#### We specify resources
#PBS -l nodes=1:ppn=2,mem=1gb
# If necessary, we include information on additional
# software, here, for example, located in $HOME/bin
export PATH=$PATH:$HOME/bin
# We move to the location of our calculation task
cd /home/$USERNAME
./helloWorld.sh
# End