• Back to Home Page our Laboratory
  

High Performance Computing: last revised on Thursday, 5th November 1998

PLEASE: in order to make sure that we can speak together giving the same meaning to the words used take one minute to read some definitions described after main text below

---

High Performance Computing

The "Technical Computing Software" package HPC2.0 provides tools for managing user-sharable resources and libraries for use in parallel message-passing programs.


We have installed the following packages:


SUNWlsf ==> LSF 2.2 licensed from "Platform Computing"
SUNWpvm3 ==> Parallel Virtual Machine PVM 3.3.11
SUNWmpi ==> Message Passing Interface MPI
SUNWrte ==> Ressource Management for parallel applications
SUNWhpf ==> High Performance Fortran Compiler
SUNWs3l ==> Sun Scientific Subroutine Library, a parallel math lib
SUNWprism ==> a graphical programming environment: enables users to debug both serial and parallel programs.

LSF means "Load Share Facility" and enables to control computing ressource allocation. It monitors the load on key system ressources and refers to a configurable usage policy.


PVM means "Parallel Virtual Machine" and MPI means "Message Passing Interface".
PVM is a first generation message-passing environment; MPI is a second generation.They are fundamentally equivalent. Both are now public domain.
RTE means "Run Time Environment"

Let us agree about some concepts:

"server" : a machine which offer some service.

Depending of the kind of service offered we differentiate between many kinds of "server":

• "cpu server": a machine which offer its CPU to run programs.
• "HPC node": is an individual machine with symmetric multiprocessors (SMPs) on which the Sun HPC software runs.
• "HPC partition": is a set of HPC nodes organized and configured with specific attributes by the Sun HPC run-time environment
• "display server": a machine on which you 'see' on its monitor the graphic output of a program which runs on a "CPU server" (can be the same machine).
• "LSF server": is a node where LSF sends jobs to be run.
• "LSF client": is a node that only submits jobs to an LSF server to run. LSF jobs do NOT run on LSF clients.
• "RTE master node": it is the host that coordinates the spawning of processes across the nodes of the cluster.
• "file server": a machine with harddisk(s) that can be 'exported' and usually mounted as NFS over the network. i.e. your data physically 'reside' on the harddisk of the "file server" but you are logged in (i.e. "working") on some other machine.

"job": a task to be executed

There are many different kinds of "jobs". I would like to define only two types: "local" and "remote".

• "local job": a task which will be executed on the "local" machine (see below).
• "remote job": a task which will executed on a "remote" machine (see below).
• "local" machine: is the machine whose keyboard, mouse and monitor you are now using. Often is the machine on your desk, sometime one machine out of a pool (as in K71).
• "remote" machine: all others   :-)

"user": a person who has a valid account on our net.

• "super user": the system manager. A trusted person who have access to all files (also system files) on a machine.

Example: Ray, Olivier, Benedikt (for UNIX machines)

• "local user": one working on a 'local' machine. Also defined as the "owner" of the 'local' machine.
• "remote user": any user who is logged into a machine via net, i.e. is not using the keyboard, mouse and monitor of the machine.

"nice": a command used to let 'remote jobs' appear nice to the 'local' user .

Thank you very much for reading all this.

Go back to the begin of this topic about HPC

Ciao Ray

P.S. Please, mail suggestions (very welcome) to ballisti@ifh.ee.ethz.ch (Click here)