
==============================================================================
MERA BENCHMARK REPORT
==============================================================================
Date         : 2026-09-07
Host         : op4pgn
CPU          : INTEL(R) XEON(R) GOLD 6534 (32 threads)
RAM          : 1511.41 GB
Filesystem   : btrfs   mount /data1
Julia        : 1.12.7, 24 compute + 24 GC threads
Mera         : 1.8.0

Dataset      : L13_SN5_CD_only output 390
  ncpu       : 5120
  levelmax   : 13   (read capped at lmax=9)
  files      : 20489
  on disk    : 53.18 GB

Reading the RAMSES output (16 threads):
  hydro      :    77.35 s +- 2.10 s
  particles  :     3.36 s +- 1.13 s
  gravity    :    35.87 s +- 2.06 s
  TOTAL      :   116.57 s

Conversion (hydro, gravity, particles, 16 threads):
  read from RAMSES   :   115.03 s
  savedata write     :     6.83 s
  one-off conversion :   121.86 s
  MERA re-read warm  :   3.45 s   (33.3x faster)
  size on disk       : 53.18 GB -> 1.7 GB   (levels up to 9 only, not comparable)
  break-even         : 1.1 re-reads
  memory to load the same data:
    allocated RAMSES :  156.71 GB
    allocated MERA   :    6.34 GB   (24.7x less churn)
    GC RAMSES / MERA : 19.25 s / 708.7 ms
    peak RSS RAMSES  :   18.16 GB
    peak RSS MERA    :    8.05 GB   (2.3x lower)

Storage by component:
  hydro        24.94 GB   46.9%
  gravity      17.52 GB   32.9%
  amr          10.66 GB   20.1%
  particle      65.0 MB    0.1%

Reading thread sweep (:hydro):
    1 threads :   256.24 s  1.00x
    2 threads :    159.6 s  1.61x
    4 threads :   113.16 s  2.26x
    8 threads :    88.26 s  2.90x
   16 threads :    75.86 s  3.38x
   24 threads :    85.87 s  2.98x
  fastest 16 threads, sweet spot 16 threads

Storage, IOPS by thread count:
    1 threads :      98933 IOPS
    2 threads :     177845 IOPS
    4 threads :     223141 IOPS
    8 threads :     254402 IOPS
   16 threads :     284662 IOPS
   24 threads :     302643 IOPS
==============================================================================
