Skip to content

Latest commit

 

History

History
69 lines (51 loc) · 3.55 KB

parallel_use.md

File metadata and controls

69 lines (51 loc) · 3.55 KB
Raw

Using the model on parallel systems

This includes (at present) shared memory systems, e.g. workstation PCs compute clusters (on a single node), HPC (on a single node)

An MPI (distributed memory) version is not yet available. (i.e. model domain spread over multiple nodes).

You will need to compile HAIL-CAESAR with a compiler that supports OpenMP. This is usually provided with most modern C++ compilers, e.g. g++/gcc, Intel, etc.

The relevant flags are provided in the Makefile. (-fopenmp (for gcc) and -DOMP_COMPILE_FOR_PARALLEL.

If the model can see a multithreaded environment, the output from STDOUT will look something like this when you run the hailcaesar executable:

##################################
#  CATCHMENT HYDROGEOMORPHOLOGY  #
#        MODEL version 1.0       #
#          (HAIL-CAESAR)         #
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

Hello! My name is OpenMP. I like to do LSD (Land Surface Dynamics) in parallel!
Your system has: 4 PROCESSORS available and 4 THREADS to use!
(Note: On some systems, threads are reported the same as processors.)
Yo, whaddup! From thread number...0
Yo, whaddup! From thread number...1
Yo, whaddup! From thread number...2
Yo, whaddup! From thread number...3
Goodbye, parallel region!

Troubleshooting

My system definitely has multiple processors/CPU cores, but I am only getting 1 thread reported?

  • You may need to explicitly set the OMP_NUM_THREADS environment variable. On Unix systems, this is done using either export OMP_NUM_THREADS=[number] (e.g. 4 for four cores). or set OMP_NUM_THREADS=[number]. You need to check with your system's documentation to see which one is necessary.

  • Usually, you should set OMP_NUM_THREADS to match the number of cores/CPUs your system has, however you may wish to experiment with a number of threads than cores if you doing a lot of heavy and frequent I/O (i.e. something crazy like writing DEMs every few seconds...)

I am confused whether my laptop/computer/server/cluster/hpc has cores/CPUs/processors/nodes or something else

  • If you have a relatively modern computer or laptop, you can almost certainly make use of the parallel options in HAIL-CAESAR. (When I'm travelling I test it on a crappy netbook, and even that has multiple cores, so it can do stuff in parallel)

  • CPU is used as a generic term nowadays to mean a processing unit. Processor can mean the physical chip(s) inside your machine (though sometimes it is used interchangeably with CPU.). Core is a region of the physical computer chip that can be treated as if it were a single, independent CPU.

  • A laptop, for example, usually has a single chip inside it, which has multiple cores. Each core can execute a portion of the HAIL-CAESAR code at the same time -- in parallel -- and so speed up the running of the model. If you were to look in the system settings, this sometimes shows up as individual CPUs.

  • To make matters more confusing, some systems (cluster computers, servers, fancy workstation computers, have multiple physical processors with multiple cores. So if you have a server with 4 processor chips, and each has 4 cores on it, you actually have 16 'CPUs' to use.

  • Nodes are something else entirely found on cluster machines and HPC/supercomputers. A node is a single unit of processors that can all access the same memory. Nodes are networked together to form supercomputers or clusters. You cannot (yet) run one instance of HAIL-CAESAR over multiple nodes, but you can run multiple instances of HAIL-CAESAR over multiple nodes, e.g. for ensemble simulations or sensitivity analyses.