Administrator Guide

Appendix A Storage array cabling
17 Dell EMC Ready Solutions for HPC BeeGFS High Capacity Storage | ID 424
The following figure shows the random read and write performance.
Random N-N reads and writes
As the figure shows, the write performance reaches around 31K IOPS and remains stable from 32 threads to
512 threads. In contrast, the read performance increases with the increase in the number of IO requests with
a maximum performance of around 47K IOPS at 512 threads, which is the maximum number of threads
tested for the solution. ME4 requires a higher queue depth to reach the maximum read performance and the
graph indicates that the performance may continue to increase with more than 512 threads. However, as the
tests were run only with 8 clients, we did not have enough cores to run more than 512 threads.
4.1.3 IOR N-1
The performance of sequential reads and writes with N threads to a single shared file was measured with IOR
version 3.3.0+dev, assisted by OpenMPI-4.0.2rc3. The benchmark was run over eight compute nodes. Tests
executed varied from single thread up to 512 threads.
We converted throughput results to GB/s from the MiB/s metrics that were provided by the tool. For thread
counts eight and above an aggregate file size of 6 TiB was chosen to minimize the effects of caching. For
thread counts below eight, the file size is 768 GiB per thread (i.e. 1.5 TiB for two threads and 3 TiB for four
threads). Within any given test, the aggregate file size used was equally divided among the number of
threads. A stripe count of 32 and transfer size of 8 MB was used.
The following commands were used to execute the benchmark for writes and reads, where $threads (the
variable for the number of threads used) was incremented in powers of two. The transfer size is 8 M and each
thread wrote to or read 128 G from a single file. Three iterations of each test have been run and the mean
value has been recorded. Figure 9 shows the N to 1 sequential I/O performance. Use the following command
to run the test:
mpirun –-allow-run-as-root -machinefile $hostlist --map-by node -np $threads
~/bin/ior -w -r -i 3 -t 8m -b $file_size -g -d 3 -e -E -k -o
$working_dir/ior.out -s 1