MPI-multicore: Applications Requirements

Questionario Originale

GAIA Einstein-Tk NEMO NAMD QuantumEpresso RegCM
1 Code Developer - E.g.: P(=personal) ,OS(=OpenSource), C(=Commercial), ...
2 Parallelization Model - E.g.: MPI, openMP, Hybrid, GPU, ..
MPI/openMP MPI/openMP MPI/openMP MPI/openMP MPI/openMP GPU MPI
3 CPU requirements - E.g.: 1-10 WholeNodes, 16-24 CPUs, ..
8-64 WholeNodes 1-64 WholeNodes 1 or more WholeNode Node 1-1024 8-128 WholeNodes 1 WholeNode
4 Memory requirements - E.g.: No, 1-2 GB per core, Memory-Bound(=whole avail. mem.), ..
Mem-bound 1GB per core 1GB per core no 1-2 GB No
5 Match Making - Yes(wms can find a suitable resource),  No( I select the resource)
No No No Yes (no preinstalled exe) No Yes
6 Compiler/Version needed on the WN - E.g.: No=compiled on the UI, gcc/g++, icc, Java ..)
gcc-c++ f90, gcc-c++ NO NO No No
7 Libraries/version needed on the WN (COMPILE time) - E.g.: no(=compiled on the UI), Blas, Lapack, Gsl, ..
cfitsio blas, lapack, gsl, hdf5, fftw3 if compiled, netcdf no no no
8 Libraries needed on the WN (EXECUTION time) - E.g.: Blas, Lapack, Gsl, ..
cfitsio blas, lapack, gsl, hdf5, fftw3 no (netcdf statically linked) no no no
9 Interpreter needed on the WN - E.g.: Python, perl, ..
no no Pyhton wrapper created No no no
10 Applications needed on the WN - Name and version (if important) of a needed Application (QE, gaussian,..)
no no no no QE-latest-vers no
11 Large amount of Data - E.g.: No, 10GB Input, 20-30 GB output, ...
400GB 20GB output 10GB input 10GB output 10 GB output 100GB
12 Data Intensive - E.g.: No(=don't care), SAN(=Shared High speed SAN), Local(=1 Local scratch disk per Node, .. )
No SAN No No SAN/Local No, files are saved to custom storage via OpenDAP
13 Expected Execution time - E .g.: No=within a few hours, 24-48h, 72h or more,..
? >3weeks >72h >3 weeks 72 or more 72 or more
14 Checkpoints - E.g.: No(=don't care) Yes(=I need a safe local storage area to save CheckPoints )
Yes on the same site Yes can continue on different sites No, output is retrieved via SE Yes, on the same site No, files are saved to custom storage via OpenDAP
15 Communication Intensive - E.g.: No (=don't care) , IB(=Infiniband needed) , ...
No IB No IB if more than 3 physical nodes used IB no
16 Web portal - E.g.: No(not needed), D(=yes for Data management), CI ( yes with Custmized Interface), CHK=(Checkpoint support)...
17 Main requirements and Expected outcomes (e.g.: optimized compilers/libraries, distributed File-systems, execution speedup, ..)
Application memory bound (up to 8TByte) - Typical resources are Supercomputers (Prace). Grid used for tests and small problems Long term execution and large file produced (>10GB). Needed fast checkpoint area on dedicated sites. Nemo needs standard HPC resources available on the Grid. The tool is statically compiled on the UI, so it doesn't need special SW env. Long execution time, but we split in several small task with checkpointing. No special Sw env neeed so match-making can be used Q/E needs to be installed (latest and optimized version) on the site. Low latency network and checkpoint needed Standard execution requires large dataset and checkpoint, but REgCM relies on custom network storage systems (openDAP). The approach based on "relocatable package" (see appl. descriptions) makes the application suitable for the Grid Environment
Topic revision: r2 - 2012-10-22 - RobertoAlfieri
This site is powered by the TWiki collaboration platformCopyright © 2008-2023 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback