lab

GDO Laboratory

This directory contains rough drafts and half-baked ideas, many of which are pseudocode, most of which are wrong. Nothing here is supposed to make sense or be useful for anything whatsoever. A very small subset may graduate from here to elsewhere in the repository.

To Do

• spike now: access controls
• how to detect changes?
  • so we can hash and distribute
  • https://upload.wikimedia.org/wikipedia/commons/3/30/IO_stack_of_the_Linux_kernel.svg
  • file layer
    • would need to re-hash entire file
    • best to just use git model: rehash if timestamp, size or attrs change
      • maybe even spot-check content
        • allow user to specify areas of interest
    • inotify
      • inotifywait -mr --format "%w%f %e" /tmp
      • XXX misses mmap write(), flush(), and close()
        • verified in testing, agrees with above diagram and with http://man7.org/linux/man-pages/man7/inotify.7.html
  • block layer
    • probably a bad idea -- a single byte insert will skew all following blocks
    • only need to re-hash changed block
      • but would need large dm block sizes
      • and would get false chunk endings when inserting and deleting bytes
    • enables simple encryption if dm-crypt on top
    • see https://docs.docker.com/engine/userguide/storagedriver/device-mapper-driver/
      • replaced aufs
      • perhaps we can base our own code on that
    • dm-log-userspace
      • allows us to run in userspace
• finish put.py
  • without chunk fault
  • with chunk fault
• finish get.py
• finish db_sqlite.py
• wire protocol is
  • chain completion protocol
  • universal chain protocol
  • universal chain completion protocol
  • microchain protocol
  • pattern completion protocol
  • sequence completion protocol
  • ordered list completion protocol
• chunk fault
  • local disk is a cache
  • cache miss causes chunk retrieval from a peer
• retrieval might be purchase from a trusted peer
  • trust builds over time
  • we rate peers based on latency
  • purchase might be in personal currency, either peer's (if we have any) or our own
• think through ultimate use of force
  • local admins can delete chunks
  • local admins can kill processes
  • local admins can filter packets
  • they need to be compensated for resource usage
  • all else is built on those foundations
• write a daemon
  • http to serve javascript GUI
  • websocket
• internode protocol
  • open connection
  • tx bid/ask list
  • rx bid/ask list, fills, and chunks
  • tx bid/ask list, fills, and chunks
  • ...
  • close connection
  • discard order book
• run rolling hash over concatenated hashes?
  • maybe with a smaller word size?
• merge cdc1, cdh1, and rough_draft_pseudocode into db_sqlite
  • turn the stream hash in cdh1 into a root hash of the hash tree; https://en.wikipedia.org/wiki/Hash_list https://en.wikipedia.org/wiki/Merkle_tree
• create a vanilla DB superclass for future storage methods
• cdch as higher-performance version of cdc1+cdh1
• reimplement cdc1p in C
• ECDSA signing
• ECDH for shared key agreement
• rename cdc1m to cdch16?