mirror of
https://github.com/meshtastic/firmware.git
synced 2025-12-24 03:31:14 +00:00
memory size debugging
This commit is contained in:
@@ -1,12 +1,22 @@
|
||||
23K + messages
|
||||
+ heap of 30ish packets, 300ish bytes per packet: 8KB
|
||||
+ heap of 70ish packets, 300ish bytes per packet: 20KB
|
||||
+ 14KB soft device RAM
|
||||
|
||||
With max length Data inside the packet
|
||||
Size of NodeInfo 104
|
||||
Size of SubPacket 272
|
||||
Size of MeshPacket 304
|
||||
|
||||
If Data was smaller: for 70 data packets we would save 7KB. We would need to make SubPacket.data and MeshPacket.encrypted into "type:FT_POINTER" - variably sized mallocs
|
||||
Size of NodeInfo 104
|
||||
Size of SubPacket 96
|
||||
Size of MeshPacket 292 (could have been much smaller but I forgot to shrink MeshPacket.encrypted)
|
||||
|
||||
therefore:
|
||||
a) we should store all ToPhone message queued messages compressed as protobufs (since they will become that anyways)
|
||||
b) shrink packet pool size because none of that storage will be used for ToPhone packets
|
||||
c) don't allocate any storage in RAM for the tophone messages we save inside device state, instead just use nanopb callbacks to save/load those
|
||||
d) a smarter MeshPacket in memory representation would save about 4KB of RAM?
|
||||
d) a smarter MeshPacket in memory representation would save about 7KB of RAM. call pb_release before freeing each freshly malloced MeshPacket
|
||||
|
||||
2000790c 00003558 B devicestate // 16KB
|
||||
2000b53c 00001000 b _cache_buffer // 4KB flash filesystem support
|
||||
|
||||
Reference in New Issue
Block a user