Here is a comparison of NeXus and FIO I/O based on the test cases of table 4.6 (5.2.2020).
Scan command: %dscan exp_dmy01 -0.1 0.1 100 0.1 Elapsed times for different measurement groups: fio nxs nxs + fio d1_c01 18.7s (46%) 19.4 (48%) 19.7 (49%) d1_c01 - d1_c04 20.2s (50%) 21.0 (52%) 21.7 (54%) d1_c01 - d1_c12 23.5s (57%) 24.6 (59%) 25.8 (61%) d1_c01, mca-2048 38.3s (74%) 21.1 (52%) 22.8 (58%) d1_c01, mca-4096 47.9s (79%) 21.6 (53%) 23.2 (56%) d1_c01, mca-8192 67.9s (85%) 21.8 (54%) 25.1 (60%) d1_c01, i-petra: 19.6s (49%) 20.5 (51%) 21.3 (53%) d1_c01 - d1_c12, mca-2048 42.2s (76%) 27.0 (63%) 28.9 (65%)
The tests displayed in the above table have been carried out to compare FIO and NeXus I/O. We see that the overhead due to NeXus is less than 10%, even if NeXus and FIO are written in parallel. We conclude that the performance reduction generated by NeXus is acceptable.