Added support for writing varying local data size in each process

[kaushikvelusamy]

Enhancement

Currently, in h5bench_write, each process writes the same amount of local data.
This program h5bench_write_var_normal_dist.c demonstrates a prototype for each process writing a varying
size local data buffer which follows a normal distribution based on the given mean number of particles provided from DIM1 and standard deviation STDEV_DIM1 in the config file.

Sample output:

$mpirun -np 16 /home/kvelusamy/h5bench-PR/gitrepos/K-h5bench/h5bench/build/h5bench_write_var_normal_dist /home/kvelusamy/h5bench-PR/gitrepos/K-h5bench/h5bench/h5bench_patterns/sample_config/sample_write_cc1d_var_normal_dist.cfg test.h5 

IO_OPERATION=WRITE
MEM_PATTERN=CONTIG
FILE_PATTERN=CONTIG
TIMESTEPS=5
DELAYED_CLOSE_TIMESTEPS=2
COLLECTIVE_DATA=YES
COLLECTIVE_METADATA=YES
EMULATED_COMPUTE_TIME_PER_TIMESTEP=1 s
NUM_DIMS=1
DIM_1=524288 # Mean number of particles for h5bench_write_var_normal_dist
STDEV_DIM_1=100000 # standard deviation for h5bench_write_var_normal_dist
DIM_2=1
DIM_3=1
CSV_FILE=output.csv


Configuration file: /home/kvelusamy/h5bench-PR/gitrepos/K-h5bench/h5bench/h5bench_patterns/sample_config/sample_write_cc1d_var_normal_dist.cfg
Output data file: test.h5

================ Benchmark Configuration ==================
File: /home/kvelusamy/h5bench-PR/gitrepos/K-h5bench/h5bench/h5bench_patterns/sample_config/sample_write_cc1d_var_normal_dist.cfg
Number of particles per rank: 524288
Number of time steps: 5
Emulated compute time per timestep: 1
Mode: SYNC
Collective metadata operations: YES
Collective buffering for data operations: YES
Number of dimensions: 1
    Dim_1: 524288
Standard deviation for varying particle size in normal distribution = 100000
===========================================================

Start benchmark: h5bench_write
Given mean particles : 524288
Given standard deviation : 100000
Total ranks 16 


Number of particles in rank 12 = 334323
Number of particles in rank 0 = 381260
 
Number of particles in rank 15 = 439010
Number of particles in rank 10 = 437476
Number of particles in rank 13 = 450079
Number of particles in rank 7 = 457785
Number of particles in rank 14 = 490708
Number of particles in rank 11 = 492040
 
Number of particles in rank 1 = 567112
Number of particles in rank 5 = 583361
 
Number of particles in rank 9 = 608116
Number of particles in rank 8 = 626028
Number of particles in rank 6 = 657237
Number of particles in rank 2 = 661096
 
Number of particles in rank 4 = 729723
Number of particles in rank 3 = 859789


Total number of particles: 8M
Collective Metadata operations: ON
Opened HDF5 file... 
Writing Timestep_0 ... 
    data_write_contig_contig_MD_array: Finished writing time step 
Computing...
Writing Timestep_1 ... 
    data_write_contig_contig_MD_array: Finished writing time step 
Computing...
Writing Timestep_2 ... 
    data_write_contig_contig_MD_array: Finished writing time step 
Computing...
Writing Timestep_3 ... 
    data_write_contig_contig_MD_array: Finished writing time step 
Computing...
Writing Timestep_4 ... 
    data_write_contig_contig_MD_array: Finished writing time step 

=================== Performance Results ==================
Total number of ranks: 16
Total number of particles: 8M
Final mean particles: 548446 
Final standard deviation: 134156.966081
Total emulated compute time: 4.000 s
Total write size: 930.811 MB
Raw write time: 19.117 s
Metadata time: 0.001 s
H5Fcreate() time: 1.639 s
H5Fflush() time: 4.299 s
H5Fclose() time: 0.786 s
Observed completion time: 29.997 s
SYNC Raw write rate: 48.688 MB/s 
SYNC Observed write rate: 35.804 MB/s
===========================================================

[jeanbez]

@kaushikvelusamy, thanks for the contribution! Before proceeding with the merge, we need a couple of additional updates:

• Could you please update the documentation accordingly with the new test and option?
• Include a sample .json configuration file in the samples folder (https://github.com/hpc-io/h5bench/tree/master/samples)
• Based on the code and tests, that option is only valid for the first dimension of the dataset, correct? Could we extend that to all 3 dimensions?
• Can you also update the CSV to output the new configuration as well? (see: https://github.com/hpc-io/h5bench/pull/88/files#diff-6e96010cf6502286f107925b40d1d971ec2d6d7a62d5852cfa6903c187317d2cR1241-R1260)

Other than that, my tests locally seem to be working. Once we have a .json file, the CI will test for all setups we have defined, including in Cori/Perlmutter.

[jeanbez]

We need to address some points in the previous comment.

[kaushikvelusamy]

updated the docs, output.csv file info, sample json file and a test file. For now, lets skip DIM2 and DIM3 support.

[jeanbez]

@sbyna, in case you want to take a second look before merging. CI is passing, and tests on Perlmutter are successful as well.