Humboldt-Universität zu Berlin - Mathematisch-Naturwissen­schaft­liche Fakultät - Experimentelle Elementarteilchenphysik

Atlas HU Cluster

Configuration and usage of the local computing cluster
 

Description:

The Atlas HU cluster consists recently of sixteen machines described in the table below. The machines are accessible by secure shell login (SSH) from every EEP desktop computer or via the gateway machines eelg01-eelg04 from outside. Each machine contains data disks collected in a RAID -5 system which is exported to all other machines of the cluster. All machines can be used as workgroup servers and are integrated into the batch system. Note that as01b because of its large RAM is the primary machine for number crunching jobs.

Machine
Data disk
Platform
CPU
#CPUs
RAM
HD size (data)
 ms01b  /rdsk/datm1
 x86_64 Intel Xeon 2.4GHz
8
 12Gb
 1.4Tb
 ms02b  /rdsk/datm2  x86_64 Intel Xeon 2.4GHz 8
 12Gb  1.4Tb
 ms03b  /rdsk/datm3  x86_64 Intel Xeon 2.4GHz 8
 12Gb  1.5Tb
 ms04b  /rdsk/datm4  x86_64 Intel Xeon 2.4GHz 8
 12Gb  1.5Tb
 ms05b  /rdsk/datm5  x86_64 Intel Xeon 2.4GHz 8
 12Gb  2.1Tb
 ms06b  /rdsk/datm6  x86_64 Intel Xeon 2.4GHz 8
 12Gb  2.1Tb
 ms07b  /rdsk/datm7  x86_64 Intel Xeon 2.4GHz 8
 12Gb  2.1Tb
 ms08b  /rdsk/datm8  x86_64 Intel Xeon 2.4GHz 8
 12Gb  2.1Tb
 as01b  /rdsk/dats1  x86_64 Intel Xeon 3.0GHz
8
 64Gb
 3.2Tb
 as02b  /rdsk/dats2  x86_64 Intel Xeon 3.0GHz 8
 64Gb  3.2Tb
 as03b  /rdsk/dats3  x86_64 Intel Xeon 3.0GHz 8
 64Gb  3.2Tb
 as04b  /rdsk/dats4  x86_64 Intel Xeon 3.0GHz 8
 64Gb  3.2Tb
 as05b  /rdsk/dats5  x86_64 Intel Xeon 3.0GHz 8
 24Gb  3.2Tb
 as06b  /rdsk/dats6  x86_64 Intel Xeon 3.0GHz 8
 24Gb  3.2Tb
 as07b
 /rdsk/dats7
 x86_64
Intel Xeon 2.4GHz
8
 24Gb
 3.2Tb
 as08b
 /rdsk/dats8
 x86_64
Intel Xeon 2.4GHz 8
 24Gb
 3.2Tb

 

Installation:

The machines are running with Scientific Linux 5 installed. The default gcc and g++ compiler versions are 4.1.2. However, the older version 3.4.6 of gcc and g++ are provided via gcc34 and g++34 resp.

Usage:

Because the computing resources are shared among an ever growing number of users all users have to follow several rules.

  • No interactive jobs beside short test jobs (0.5-1h) are allowed !
  • All jobs must be submitted to the batch system
  • Before putting load to any machine either by interactive work or batch jobs make sure not to overload the machine. Check if the machine in question is already running other jobs. Monitor the CPU and memory load of the machine during processing. In case you do not know how to achieve all this get yourself acquainted with tools like top, xload, ps and kill. In case of running interactively it is always good practice to renice your jobs.
  • Try to avoid memory leaks in your code. Test your jobs thoroughly in this respect before submitting a large number of jobs. Jobs with large memory load not only slow down the system, in the end the lead to endless swapping which reduces much the life-time of the system disks.
  • Do not use these machines for editing files (emacs), web browsing or email clients! Such things should be done on the desktop machines or, when working from outside, at the gateway machines eelg01-04.
  • Always keep in mind that you share the machines with others who may need the resources at least as badly as you need them. Be polite.