Benchmark Results

Node Status Output

During distributed execution, the status of each MPI rank was written to disk for later inspection. This was accomplished by exporting rank output to a shared NFS directory accessible by all nodes.

The following option was added to the mpirun command to enable output collection:

--output-filename /mnt/nfs_share/status

This option designates a directory where each MPI rank writes its execution status and performance data. The collected output allows individual node performance to be examined after execution completes.

The results were retrieved from the shared directory using the following command:

cat /mnt/nfs_share/status/*

Cluster Performance Results

In a test run, per-rank statistics were written to a file named rankstats.txt. Aggregating the output from all ranks produced the following total throughput for the eight-node CPU cluster:

• Total passwords tested per second: 253,485.512 p/s
• Total hash computations per second: 5,076,790.614 c/s

GPU Performance Results

For comparison, the same password set was processed on a standalone system equipped with a single Radeon RX 6600 GPU. The following performance metrics were observed:

• Passwords tested per second: 207,813 p/s
• Hash computations per second: 7,936,000 c/s

Summary

The distributed CPU cluster demonstrated strong aggregate password-testing throughput when scaling across eight nodes. While the GPU achieved a higher raw hash computation rate, the cluster provided competitive performance through parallel execution and horizontal scaling.