Skip to content

hpc-carpentry/amdahl

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

36 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

"Black Box" Amdahl's Law Simulator

This Python module contains a pseudo-application that can be used as a black box to reproduce Amdahl's Law. It does not do real calculations, nor any real communication, so can easily be overloaded.

The application is installed as a Python module with a shell script wrapper. The only requirements are a Message Passing Interface (MPI) library and the MPI for Python module, mpi4py.

Background

Amdahl's Law posits that some unit of work comprises a proportion $p$ that benefits from parallel resources, and a proportion $s$ that is constrained to execute in serial. The theoretical maximum speedup achievable for such a workload is

$$ S = \frac{1}{s + p/N} $$

where $S$ is the speedup relative to performing all of the work in serial and $N$ is the number of parallel workers. A plot of $S$ vs. $N$ ought to look like this, for $p = 0.89$ (since $s = 1 - p = 0.11$):

  5┬────────┼────────┼────────┼────────·────────┼────────┼────────┼────────┼────────*
   │                                  ·                                             │
   │                                 ·                                     *        │
   │                                ·                                               │
   │                               ·                              *                 │
   │                              ·                                                 │
   │                             ·                                                  │
   │                            ·                        *                          │
   │                           ·                                                    │
  4┼                          ·                                                     ┼
   │                         ·                  *                                   │
   │                        ·                                                       │
   │                       ·                                                        │
   │                      ·                                                         │
   │                     ·             *                                            │
S  │                    ·                                                           │
p  │                   ·                                                            │
e  │                  ·                                                             │
e 3┼                 ·        *                                                     ┼
d  │                ·                                                               │
u  │               ·                                                                │
p  │              ·                                                                 │
   │             ·                                                                  │
   │            ·    *                                                              |
   │           ·                                                                    │
   │          ·                                                                     │
   │         ·                                                                      │
  2┼        ·                                                                       ┼
   │       ·                                                                        │
   │      · *                                                                       │
   │     ·                                                                          │
   │    ·                                                                           │
   │   ·                                                                            │
   │  ·                                                                             │
   │ ·                                                                              │
   │·                                                                               │
  1*────────┼────────┼────────┼────────┼────────┼────────┼────────┼────────┼────────┤
   1        2        3        4        5        6        7        8        9       10
                                         Workers

"Ideal scaling" ($p = 1$) is represented by the dotted line, $S = N$.

This graph shows there is a speed limit for every workload, and diminishing returns on throwing more parallel processors at a problem. It is worth running a "scaling study" to assess how far away that speed limit might be for the given task.

Setup

An external requirement is a Message Passing Interface (MPI) library, such as OpenMPI or MPICH. To check whether you have one installed, run

which mpirun

If the command returns a filesystem path, e.g. /usr/bin/mpirun, you're all set! If not, contact an administrator or use your operating system's package manager to install both the binaries and the development headers for either OpenMPI or MPICH.

Once an MPI library is installed, you should be able to install the amdahl package and its dependencies using pip:

python3 -m pip install --user amdahl

Usage

Once installed, you can run the program using

mpirun -np 2 amdahl

You should see something like the following output:

Doing 30 seconds of 'work' on 2 processors, which should take 18 seconds
with 0.8 parallel proportion of the workload.

  Hello, World! I am process 0 of 2. I will do all the serial 'work' for 6 seconds.
  Hello, World! I am process 0 of 2. I will do parallel 'work' for 10 seconds.
  Hello, World! I am process 1 of 2. I will do parallel 'work' for 14 seconds.

Total execution time (according to rank 0): 16 seconds