Differenze

Queste sono le differenze tra la revisione selezionata e la versione attuale della pagina.

--- roberto.alfieri:pub:mpigrid [25/02/2010 21:19] – roberto.alfieri
+++ roberto.alfieri:pub:mpigrid [26/02/2010 12:10] (versione attuale) – roberto.alfieri
@@ Linea 1: / Linea 1: @@
+=====  MPI/theophys Project ====
+==== WorkerNodes ====
+== Software environment ==
+   * SL5.x x86_64
+   * openMPI >= 1.3 (1.4 would be better)
+   * MPICH2
+   * Gnu C, C++, Fortran (gfortran, g77 , g90??)
+   * Support for commercial compilers?
+   * Scientific libraries:
+      * openMP: multithread library
+      * HDF5  : Data storing and managing library.
+      * Blas : Basic Linear Algebra Subprograms
+      * lapack: Linear Algebra PACKage
+      * GSL : Gnu Scientific Library
+      * GMP:  Gnu Multiprecision Library
+      * GLPK: Gnu Linear Programming Kit.
+      * Fftw3: Fast Fourier Transform Library
+      * Octave : high-level language for numerical computations
+==Installation example==
+<code>
+ yum install -y yum-conf-epel.noarch
+ yum install -y octave hdf5-devel glpk fftw3
+ yum install -y libgomp blas-devel gsl-devel gmp-devel
+</code>
+== theophys TAG==
+The following is a possible TAG to be published by theophys compliant sites:
+  GlueHostApplicationSoftwareRunTimeEnvironment: VO-theophys-gcc41 ??
+====  Cluster ====
+== Published TAGs for MPI ==
+Mpi-start is the way to start MPI jobs:
+   MPI-START
+Al least openMPI should be installed:
+   MPI_OPENMPI
+   MPI_OPENMPI_VERSION="x.y.z"
+Shared home is recommended, but file distribution is supported by MPI-start:
+   MPI_SHARED_HOME | MPI_NO_SHARED_HOME
+Remote Start-up of MPI job can be achieved via password-less SSH:
+  MPI_SSH_HOST_BASED_AUTH
+Infiniband is recommended, but Gbit (or 10Gb) Ethernet can be used:
+    MPI-Infiniband | MPI-Ethernet
+==Open Issues==
+  * Is it possible to publish the actual number  of free CPUs per queue?
+  * How is CpuNumber used in the match-making process?
+<code>
+At the moment  CpuNumber is not used at all for match making.
+Temporary solution in the JDL:
+CPUNumber=n
+other.GlueCEInfoTotalCPUs >= CPUNumber
+</code>
+==== JDL ====
+==Typical parallel jdl==
+  JobType = "Normal" ;
+  CpuNumber = 8 ;
+==multithread support ==
+  SMPGranularity = 8;
+  WholeNodes = True;
+Multithread support is desirable and it should be integrated
+ in the middleware as soon as possible.
+== Open Issues ==
+  * Is it possible to integrate Granularity/WholeNodes directly in InfnGrid?
+<code>
+CREAM and BLAH: see https://twiki.cern.ch/twiki/bin/view/EGEE/ParameterPassing ??
+WMS: included in WMS 3.3
+</code>
+==== Parallel and sequential jobs ====
+VOMS Roles can be used to limit the access to Parallel queues.
+== Voms Role = "Parallel" ==
+The [[https://voms.cnaf.infn.it:8443/voms/infngrid/SearchRoles.do | Role]] is assigned by the VO manager
+and released by [[https://voms.cnaf.infn.it:8443/voms/theophys/Siblings.do |VOMS]] only on explicit request.
+==Setup example==
+<code>
+site-info.def:
+PARALLEL_GROUP_ENABLE="/infngrid/ROLE=parallel"
+/opt/glite/yaim/defaults/ig-site.pre:
+FQANVOVIEWS=yes
+groups.conf:
+"/infngrid/ROLE=parallel"::::
+voms-proxy-init -voms infngrid:/infngrid/Role=parallel
+voms-proxy-info -all
+>....
+>attribute : /infngrid/Role=parallel/Capability=NULL
+>attribute : /infngrid/Role=NULL/Capability=NULL
+>...
+</code>
+==== MPI multi-thread jobs ====
+MPI and multi-thread programs can be combined to exploit
+the upcoming multicore architectures.
+The hybrid mutlithread/MPI programming leads to  a request
+of N CPUs with a smaller number of MPI processes (N/thread_num).
+Actually this programming model is not supported in EGEE.
+Possible solution:
+modify  the value type of WholeNodes from boolean to integer.
+Example:
+  SMPGranularity = 8;
+  WholeNodes = 4;
+This syntax would lead to
+  qsub -l nodes=4:ppn=GlueHostArchitectureSMPSize
+where ppn is a number >=8.
+WholeNodes value should be passed to mpi-start
+as the number of MPI processes.
+Mpi-start should be  modified accordingly.
+Mixed mpi/mutithread programs require thread safe
+ MPI implementations.
+Thead safety can be easily  verified:
+<code>
+MPI_Init_thread(&argc, &argv, 3, &prov);
+printf("MPI_Init_thread provided:%d\n", prov);
+</code>
+The third parameter (number 3)  means a request
+of full thread safety support ( MPI_THREAD_MULTIPLE ).
+If returned value for prov is 0  thread support is not provided
+(MPI_THREAD_SINGLE).
+==== Scheduling ====
+==objectives==
+  * Minimize jobs starvation
+  * Maximize resources exploitation
+==Possible scenario==
+MPI sites with at least 2 queues sharing the same pool of WNs:
+  * **high priority parallel queue**
+      * accessible only with special Role (Role=parallel ?)
+  *  **Low priority sequential queue**
+      * preemptable (renice or requeue ?)
+      * short WallClockTime (less than 6 hours?)
+      * accessible only with special Role (Role=short ?).
+==== Revision history ====
+  * 20100225 - R. DePietri,  F. DiRenzo -  User's required libraries
+  * 20100210 - C. Aiftimiei, R.Alfieri, M.Bencivenni, T.Ferrari - First Version