Mpirun on Rocks clusters is used to launch jobs that are linked with the Ethernet device for MPICH.
You must run HPL as a regular user (that is, not root). If you don’t have a user account on the cluster, create one for yourself with:
# useradd username
For example, to interactively launch the benchmark “High-Performance Linpack” (HPL) on two processors:
* Create a file in your home directory named machines, and put two entries in it, such as:
compute-0-0
compute-0-1
* Download the the two-processor HPL configuration file and save it as HPL.dat in your home directory.
* Now launch the job from the frontend:
$ /opt/mpich/gnu/bin/mpirun -nolocal -np 2 -machinefile machines /opt/hpl/gnu/bin/xhpl
Bé, els problemes que tenia eren:
a) Quan es crea l’usuari i després entres al seu compte no se li ha d’especificar contrasenya per a la clau ssh, ja que sinó quan envies a executar els processos es queden atrapats en l’autorització.
b) He de probar si els programes es poden enviar a executar no des de dins el frontend, sinó des d’algun node que disposi d’openMosix, ja que sinó no podré comprovar el rendiment d’MPI a través d’openMosix.
c) Continuo tenint un problema, l’HPL no es propaga cap a un dels nodes. Si al fitxer machines poso el compute-0-1 primer l’aplicació no es llança, i si el poso després només propaga cap al compute-0-0. Vaia coses més rares.
Més coses interessants, ara toca accedir des de màquines exteriors al servidor web (tocant iptables, per supost):
3.1.2. Enabling Public Web Access with Control Lists
To permenantly enable selected web access to the cluster from other machines on the public network, follow the steps below. Apache’s access control directives will provide protection for the most sensitive parts of the cluster web site, however some effort will be necessary to make effective use of them.
HTTP (web access protocol) is a clear-text channel into your cluster. Although the Apache webserver is mature and well tested, security holes in the PHP engine have been found and exploited. Opening web access to the outside world by following the instructions below will make your cluster more prone to malicious attacks and breakins.
1. Edit the /etc/sysconfig/iptables file. Uncomment the line as indicated in the file.
…
-A INPUT -i eth1 -p tcp -m tcp –dport ssh -j ACCEPT
# Uncomment the line below to activate web access to the cluster.
#-A INPUT -i eth1 -p tcp -m tcp –dport www -j ACCEPT
… other firewall directives …
2. Restart the iptables service. You must execute this command as the root user.
$ service iptables restart
3. Test your changes by pointing a web browser to http://my.cluster.org/, where “my.cluster.org” is the DNS name of your frontend machine.
If you cannot connect to this address, the problem is most likely in your network connectivity between your web browser and the cluster. Check that you can ping the frontend machine from the machine running the web browser, that you can ssh into it, etc.
També he provat a executar tasques amb el PBS, seguint aquestes indicacions. Per als treballs en serie tot ha funcionat com s’esperava, però per als treballs en paral lel no se q passa q no hi ha forma de q treballin.
El deixaré tota la nit treballant, pa que disfrute xD
Recent Comments