CLUSTER BENCHMARK WORKSHEET — September 7, 2026
Guide: https://yenra.com/cluster-computing/
Choose strong scaling (fixed complete job) or weak scaling (fixed work per
worker). Record the resource unit: identical nodes, processes, or cores.
Use comparable configurations. This worksheet is a record, not a scheduler.

Task / input version or hashes / output acceptance criterion:
Software / compiler / libraries / options:
Resource unit / node hardware / cores allocated / memory / network / storage:
Scaling type / problem size:
Baseline elapsed execution time and units:
For each run record:
  Resources N:
  Repetition / date / competing load / cache state:
  Queue time / execution time T_N / time to usable result:
  Correctness check / result:
  Speedup = baseline execution time / T_N:
  Efficiency (%) = 100 * speedup / N (baseline uses one resource unit):
  Resource-time = N * T_N (state units):
  Memory / communication / I/O observations:
  Failure or retry handling:

FICTIONAL FIXED-JOB EXAMPLE (each node has the same allocated cores)
Nodes  Minutes  Speedup  Efficiency  Node-minutes
1      100      1.00     100.0%      100
2       58      1.72      86.2%      116
4       34      2.94      73.5%      136
8       25      4.00      50.0%      200
Speedup and efficiency rounded for display, calculated from original times.
All example outputs assumed correct. No real equipment measured here.
Node-minutes are not money without a stated charging model and rate.

Idealized serial-fraction example: 10% serial, 90% perfectly parallel gives
time fraction = 0.10 + 0.90/N; at N=8, speedup=4.705882... (4.71 rounded).
Excludes communication/other overhead; not a hardware performance prediction.

Interpretation / fastest acceptable configuration / resource tradeoff:
Changes to investigate / next controlled test:
Sources: https://hpc.llnl.gov/documentation/tutorials/introduction-parallel-computing-tutorial
https://slurm.schedmd.com/job_array.html
https://www.mpi-forum.org/docs/