sequoiar / tripled Goto Github PK

Introduction:

        This is a very simple distributed file system borrowing many ideas from
        GFS.  The idea is to focus on simplicity and performance.  Not
        availability or fault tolerance (at all).

        Triple D:
                1) Dead simple...
                2) Done right...
                3) Distributed file system

Why:

        Tired of bullshit configurations and degraded performance?
        Want good performance?
        How about usable and understandable code?

Requirements:

        zeromq-2.1.7
        redis-2.2.7

Usage:
        master.py       -- run on metadata master node
        worker.py       -- run on worker nodes

        cptoddd.py      -- use to copy files into TripleD
        cpfromddd.py    -- use to copy files out of TripleD
        lsddd.py        -- use to list files in TripleD

Design:

        1) Master handles all metadata
        2) Workers are dumb trusting writes from clients
        3) Clients do RPC with master and workers directly
        4) Metadata comes from master, flows via client to workers
        5) Assume write once, read many in future; no append; no seek
        6) Client allowed to arbitrarily chunk (client knows best!)
        7) No replication right now; client would get list from master
        8) No multi-reading or multi-writing...
        9) All workers store chunks in the same path on their system
        10) Dumb round-robin assignment of chunks; worst case first node gets
            more data; last node gets least data
        11) Variable-sized chunks allowed
        12) No delete or rewrite


TODO:
       1) master should only write into Redis on receipt of write info from a
          worker
       2) concurrent writing to multiple nodes, reading from multi nodes
          (readahead as well)---basically add threading with push/pull sockets