matthew
05fb2709a5
iter5 (256-frame channel) regressed -1.8 MB/s vs iter4. Reverting to 32 frames. iter4 sample pattern shows clear ~30 s oscillation (peak 45 MB/s → dip 3 MB/s → recovery). Matches Linux vm.dirty_expire_centisecs default (30 s). 32 MiB chunks at 25 MB/s issue WAIT_AFTER every ~1.3 s, which can't outrun the kernel's own page-age limit, so pages buildup then flush in bursts. Smaller 8 MiB chunks (WAIT_AFTER every ~0.33 s) should keep the dirty-page set young and eliminate the periodic flush-burst dip.
agent-a1b258ca2bea02faf
@ aa536f6ab3
agent-a1cbe2cec72727536
@ f952bcfce9
agent-a82b8fb113069acd9
@ dcde016bd3
agent-ae58900dbeb230583
@ ccf7d97640
agent-aee3289bc097d110f
@ 0d4879948b
agent-afd076fb7145099f7
@ 5950156cf0
halt-token
@ 8693add66c
pes-source-sink
@ 919096b67f
pipeline
@ 8e7d1bea96
round1-fixes
@ 78b50dd5a6
round2-decrypt-decorator
@ 9e51ee9946
round2-discstream-source
@ f10c83ffe4
round2-framesink
@ eb04fdaffa
round2-halt
@ 3ca63235e0
sector-source-sink
@ 8c592d08e9
writeback-file-rename
@ e5a32a8f16