Royal Institute of Technology
School of Biotechnology
Department of
Theoretical Chemistry

Computing resources

Theoretical Chemistry Home Page > Documentation

Currently the resources are divided into three groups:

Local machines running PBS.
Supercomputers and clusters.
Standalone machines without queuing system.
Other Linux and Windows computers.

PBS machines

Report any problems or questions to the support mail.

Most of the computers in our department are connected to a queuing system, torque. The computers are using Linux OS and have the following specifications:

Label¹	Processors	Memory	Disk to run in	#m ²
C, p4	dual-processor 3.0 GHz Intel Xeon	2 GB	fast disks in dir: /scratch	11
D, p4	dual-processor 2.8 GHz Intel Xeon	2 GB	/tmp	12
E, opteron	dual-processor 2.2GHz AMD Opteron 64-bit architecture	3 GB [9 GB (etorofu.priv)]³	/tmp	7

1)	In your PBS script, use `#PBS -l nodes=p4` for C and Dmachines (default at the moment). Use `#PBS -l nodes = opteron` for E-machines. See below for more options (Requesting particular architectur).

2)	#m = nr of machines available.
3)	etorofu requested by `#PBS -l nodes=big`

Memory guidelines: Each node runs in general 2 jobs. This sets limits on the maximum memory you can specify in the input. Please do not ask for more than %mem = 500mb on D machines, 800 mb on C-machines and 1300 mb on E-machines. If you need more than that, ask for two procssors in your script file, for example like this: #PBS -l nodes=D:ppn=2

Working with PBS

All jobs must be submitted (and queried) from omega.biotech.kth.se with qsub program: qsub pbs_script (see PBS script-examples.)
To see the particulars of your running job (on what machine it is running, job id, requested time, etc.), you can use the flag -n:
```
qstat -n <Job Id>
```
Type man qstat to obtain more information on different flags of this command, and what they do.
When you know where your job is running, you can log into the node where the job is and check the output in the directory /tmp (D and E machines) or /scratch (C machines).
A job can be removed from the queue by using qdel:
```
qdel <Job Id> 
```
A list of all machines in the queue system and their status can be obtained with
```
pbsnodes -a
```
It is not possible to run parallel jobs on these machines - instead use some of the clusters we have access to.

Queues available

Queue Name	Description
Testque	For jobs up to 5 min.
shortque	For jobs taking 5 min to 4h.
longque	For jobs between 4h and 50h
verylong	For jobs between 50h and 600h.
Equeue	Opteron jobs only. Up to 600h.

Jobs are sorted into the different queues depending on the time they ask for in the script file. Sorting to the Equeue occurs only if the following is added to the script: #PBS -q Equeue

Only the C-machines have scratch directory. Since /scratch is normally faster than /tmp you should initalize your working directory as /scratch, if it is present, adding to your script something like this:
```
SCR_DIR=/scratch; [ -d $SCR_DIR] || SCR_DIR=/tmp
```
The syntax depends on the shell you are using, so be careful.

Requesting particular architecture: If your program can run on any of our machines y\ou may request:
```
#PBS -l nodes=1 
```
If you want an to use an Opteron machine (64-bit), add the following to your script:
```
#PBS -l nodes=opteron
```
If you want a C or D machine (this is the default!) use:
```
#PBS -l nodes=p4
```
If you want C/D/E-machine specifically, use this:
```
#PBS -l nodes=C (or D, or E)
```
For instance, if you know you are running a calculation which is doing a lot of I/O you need /scratch to be present so you want to use C-machines only. If you need a lot of scratch space, you can run your job on cesium, which has 90 GB scratch by using:
```
#PBS -l nodes=bigdisk 
```
If you need a lot of memory (see also above for memory guidelines), ask for both processors on a node, for example like this:
```
#PBS -l nodes = D:ppn=2 
```
There is one E machine with 9GB memory (etorofu), you can request this machine using:
```
#PBS -l nodes=big 
```

#PBS -q Equeue

Supercomputers and clusters

Information regarding our Lenngren cluster.

We also have allocated time on the following machines:

Monolith and Ingvar at NSC, Linköping - Dalton, Gamess and Gaussian 98/03
SGI3800 at NSC, Linköping - Dalton (in /nsc/home/x_oscru/dft-qr/dft-qr/dalton), Gaussian 98/03, MatLab
Lucidor at PDC, Stockholm - Dalton, Gamess, Jaguar and Gaussian 98/03
Seth at HP2CN, Umeå - Dalton (in /kfs/home/o/orubiop2/Dalton)

If you want to use them, first you should obtain an account. Go to the relevant website to find out how this could be done.

Standalone machines

As pretty much all of our desktop machines are connected to the queue, you may submit only short jobs using "nice +18", after checking that there aren't too many jobs running already. Observe that there are no more than 2 calculations per machine (1 calculation per processor). You can check that by typing top. Example execution:
```
nice +18 ./a.out &
```
Notice, that depending on what shell you use, this command may slightly differ. Particularly if you use bash, it takes the form:
```
nice -18 ./a.out &
```

Other Linux and Windows computers

There are also a few desktop computers that do not fullfill the reqirements to be in the queueing system. All computers in the department are using Linux OS except:

In the department there is also a Windows computers for common use. Here you can prepare MS Powerpoint presentation, and use SciFinder. If you find out the geographical location of this machine, please mail webmaster-at-theochem so I can add this information!

| Theoretical Chemistry Home Page > Documentation

webmaster-at-theochem.kth.se