Byte-identity proof of the memory reductions. Written by tools/proof_report.py. What each run is (the RUN file in its folder is the exact command, env.txt its environment): A run_controller.py unchanged, build N, pilot env, pynn_seed 5, read-only observer B run_controller_mem.py (R0, R1, R3, R4), build NM (R2), OPENBLAS_NUM_THREADS=4, pynn_seed 5, read-only observer C as B, pynn_seed 6: the control, must differ D as B without the observer, started without MallocSpaceEfficient so the wrapper re-executes itself E as D on build N: the wrapper without R2, run straight after D for a fair speed comparison of R2 D2 repeat of D, straight after E B2, C2, D3 as B, C, D with the final wrapper, which adds R10 (the NodeCollection leak fix) B, C, D, E, D2 ran the wrapper as in runs/short/run_controller_mem.v1-as-run.py (5ceef7d9...), without R10. One earlier attempt at C stopped at 120 s by a tool time limit: runs/proof/C-attempt1. The first smoke tests are runs/proof/smoke*. Sheet: sx 600 um, LGN (1.6,1.6) deg, or_map_stretch 27.916667; 4 NEST threads; runs/proof/sequence-proof.json (3 steps, one a plaid), lead 504 ms. reference A: runs/proof/A A: runs/proof/A connections 802912, kernel num_connections 802912, projections 16, LGN cells [('X_OFF', 512), ('X_ON', 512)], samples [225, 225] sheet: trains, spikes over all segments: {"V1_Exc_L2/3": [2160, 14379], "V1_Exc_L4": [2160, 8344], "V1_Inh_L2/3": [540, 18328], "V1_Inh_L4": [540, 6794], "X_OFF": [2048, 14782], "X_ON": [2048, 24524]} segment files 24 B: runs/proof/B connections 802912, kernel num_connections 802912, projections 16, LGN cells [('X_OFF', 512), ('X_ON', 512)], samples [225, 225] sheet: trains, spikes over all segments: {"V1_Exc_L2/3": [2160, 14379], "V1_Exc_L4": [2160, 8344], "V1_Inh_L2/3": [540, 18328], "V1_Inh_L4": [540, 6794], "X_OFF": [2048, 14782], "X_ON": [2048, 24524]} segment files 24 C: runs/proof/C connections 800523, kernel num_connections 800523, projections 16, LGN cells [('X_OFF', 512), ('X_ON', 512)], samples [225, 225] sheet: trains, spikes over all segments: {"V1_Exc_L2/3": [2160, 15283], "V1_Exc_L4": [2160, 9543], "V1_Inh_L2/3": [540, 21439], "V1_Inh_L4": [540, 7829], "X_OFF": [2048, 14694], "X_ON": [2048, 25028]} segment files 24 D: runs/proof/D sheet: trains, spikes over all segments: {"V1_Exc_L2/3": [2160, 14379], "V1_Exc_L4": [2160, 8344], "V1_Inh_L2/3": [540, 18328], "V1_Inh_L4": [540, 6794], "X_OFF": [2048, 14782], "X_ON": [2048, 24524]} segment files 24 E: runs/proof/E sheet: trains, spikes over all segments: {"V1_Exc_L2/3": [2160, 14379], "V1_Exc_L4": [2160, 8344], "V1_Inh_L2/3": [540, 18328], "V1_Inh_L4": [540, 6794], "X_OFF": [2048, 14782], "X_ON": [2048, 24524]} segment files 24 D2: runs/proof/D2 sheet: trains, spikes over all segments: {"V1_Exc_L2/3": [2160, 14379], "V1_Exc_L4": [2160, 8344], "V1_Inh_L2/3": [540, 18328], "V1_Inh_L4": [540, 6794], "X_OFF": [2048, 14782], "X_ON": [2048, 24524]} segment files 24 B2: runs/proof/B2 connections 802912, kernel num_connections 802912, projections 16, LGN cells [('X_OFF', 512), ('X_ON', 512)], samples [225, 225] sheet: trains, spikes over all segments: {"V1_Exc_L2/3": [2160, 14379], "V1_Exc_L4": [2160, 8344], "V1_Inh_L2/3": [540, 18328], "V1_Inh_L4": [540, 6794], "X_OFF": [2048, 14782], "X_ON": [2048, 24524]} segment files 24 C2: runs/proof/C2 connections 800523, kernel num_connections 800523, projections 16, LGN cells [('X_OFF', 512), ('X_ON', 512)], samples [225, 225] sheet: trains, spikes over all segments: {"V1_Exc_L2/3": [2160, 15283], "V1_Exc_L4": [2160, 9543], "V1_Inh_L2/3": [540, 21439], "V1_Inh_L4": [540, 7829], "X_OFF": [2048, 14694], "X_ON": [2048, 25028]} segment files 24 D3: runs/proof/D3 sheet: trains, spikes over all segments: {"V1_Exc_L2/3": [2160, 14379], "V1_Exc_L4": [2160, 8344], "V1_Inh_L2/3": [540, 18328], "V1_Inh_L4": [540, 6794], "X_OFF": [2048, 14782], "X_ON": [2048, 24524]} segment files 24 A against B: equal of compared lgn_input 8 of 8 connections source 17 of 17 connections target 17 of 17 connections weight 17 of 17 connections delay 17 of 17 connections address 16 of 16 connections receptor 17 of 17 connections tsodyks_parameters 16 of 16 connections final_state 16 of 16 projection names 17 of 17 recorder links 1 of 1 spike trains by segment 24 of 24 export 1 of 1 A against C: equal of compared lgn_input 4 of 8 connections source 1 of 17 connections target 1 of 17 connections weight 1 of 17 connections delay 1 of 17 connections address 0 of 16 connections receptor 1 of 17 connections tsodyks_parameters 0 of 16 connections final_state 0 of 16 projection names 17 of 17 recorder links 1 of 1 spike trains by segment 4 of 24 export 0 of 1 A against D: equal of compared spike trains by segment 24 of 24 export 1 of 1 A against E: equal of compared spike trains by segment 24 of 24 export 1 of 1 A against D2: equal of compared spike trains by segment 24 of 24 export 1 of 1 A against B2: equal of compared lgn_input 8 of 8 connections source 17 of 17 connections target 17 of 17 connections weight 17 of 17 connections delay 17 of 17 connections address 16 of 16 connections receptor 17 of 17 connections tsodyks_parameters 16 of 16 connections final_state 16 of 16 projection names 17 of 17 recorder links 1 of 1 spike trains by segment 24 of 24 export 1 of 1 A against C2: equal of compared lgn_input 4 of 8 connections source 1 of 17 connections target 1 of 17 connections weight 1 of 17 connections delay 1 of 17 connections address 0 of 16 connections receptor 1 of 17 connections tsodyks_parameters 0 of 16 connections final_state 0 of 16 projection names 17 of 17 recorder links 1 of 1 spike trains by segment 4 of 24 export 0 of 1 A against D3: equal of compared spike trains by segment 24 of 24 export 1 of 1 SHA-256 over all segment spike hashes, per run: A d0f6a8363dcd5131ce3f81e3a6b2cc38f3f49e086b091a8d5a4e06d012b53681 B d0f6a8363dcd5131ce3f81e3a6b2cc38f3f49e086b091a8d5a4e06d012b53681 C c94f71dc70788cec038b036c14ef97b484b810ebc8eacc33cdbb127be10759b9 D d0f6a8363dcd5131ce3f81e3a6b2cc38f3f49e086b091a8d5a4e06d012b53681 E d0f6a8363dcd5131ce3f81e3a6b2cc38f3f49e086b091a8d5a4e06d012b53681 D2 d0f6a8363dcd5131ce3f81e3a6b2cc38f3f49e086b091a8d5a4e06d012b53681 B2 d0f6a8363dcd5131ce3f81e3a6b2cc38f3f49e086b091a8d5a4e06d012b53681 C2 c94f71dc70788cec038b036c14ef97b484b810ebc8eacc33cdbb127be10759b9 D3 d0f6a8363dcd5131ce3f81e3a6b2cc38f3f49e086b091a8d5a4e06d012b53681 A against B identical in every item: True A against D identical in every item (spike trains, export): True C differs from A in: ['connections address', 'connections delay', 'connections final_state', 'connections receptor', 'connections source', 'connections target', 'connections tsodyks_parameters', 'connections weight', 'export', 'lgn_input', 'spike trains by segment'] C differs in connections and spike trains: True VERDICT B, C, D: PASS A against B2 identical in every item: True A against D3 identical in every item (spike trains, export): True C2 differs from A in: ['connections address', 'connections delay', 'connections final_state', 'connections receptor', 'connections source', 'connections target', 'connections tsodyks_parameters', 'connections weight', 'export', 'lgn_input', 'spike trains by segment'] C2 differs in connections and spike trains: True VERDICT B2, C2, D3: PASS A against E identical in every item: True A against D2 identical in every item: True VERDICT: PASS Speed, seconds. Only Simulator run time and connector time compare fairly between A and B: A uses 8 OpenBLAS threads and B 4, which sets the LGN step. run total simulator mozaik connectors stimulus memcap s peak GiB run_seconds A 1610 44 1550 11 1580+29 1638 1.122 1630.6 B 156 38 96 10 115+40 183 0.753 176.5 C 67 18 37 3 44+23 85 0.698 80.0 D 83 20 48 3 52+31 95 0.344 89.6 E 124 27 72 7 82+41 138 0.462 132.4 D2 251 48 172 15 193+57 276 0.379 269.4 B2 134 23 79 3 73+61 162 0.595 156.4 C2 256 48 185 14 217+39 291 0.612 284.2 D3 178 41 109 9 127+50 196 0.361 189.8