GitHub
Moderator
[SoftCSA] Restore original recording buffer size to reduce LowMem pressure
The ScrambledThread used a fixed buffer of 256*188 (47 KB) per slot,
while the original FileThread default was packetsize*1024 (188 KB).
This 4x reduction caused ~4x more write() syscalls, increasing
kernel-side allocations (NFS RPC structs, socket buffers, slab objects)
in the DMA/LowMem zone. On heavy recording setups (4+ simultaneous
NFS recordings), this contributed to LowMem exhaustion and OOM kills.
Passing -1 restores the default buffer calculation. The descrambling
code is unaffected as it iterates 188-byte packets regardless of
buffer size.
Continue reading...
The ScrambledThread used a fixed buffer of 256*188 (47 KB) per slot,
while the original FileThread default was packetsize*1024 (188 KB).
This 4x reduction caused ~4x more write() syscalls, increasing
kernel-side allocations (NFS RPC structs, socket buffers, slab objects)
in the DMA/LowMem zone. On heavy recording setups (4+ simultaneous
NFS recordings), this contributed to LowMem exhaustion and OOM kills.
Passing -1 restores the default buffer calculation. The descrambling
code is unaffected as it iterates 188-byte packets regardless of
buffer size.
Continue reading...