Venti/VT

indirect block 3 May

array of blockrefs

blockref 3 May

bs(flags:indirect)bs(span)bs(hashen)hash

.vt file content 2 May

each line:
hex(dirent) iso8601time comment(username direntName etc…)

dirent format 2 May

bs(type:0=file,1=dir,...) bs(flags:1=hasmeta,2=hasname) bs(metalen)meta blockref

TODO

writable files – each write sequence consists of a “file” overlaid on the read-only previous content

http interface

fuse interface

inodes/hard links?

symlinks

class diagram: store

block interface: either a blockref or indirect blockref or a “live file” later

iblock encoding: flags, size, hash; flags=(indirect, hasmeta, oddhash)

webdav interface once fs working?

9p server?

rolling hash in C; yielding (offset,hash) tuples?

Prefetch multiple blocks, eg from an iblock. Scatter/gather wrapper? With progressive yield of blocks in order

“vt push targetStore”

“vt scan” – don’t open -S/-C stores

sync() on stream/tcp close

flat file cache for blocks? make temp unlinked, trunc, append, rewind at threshold

optional compression in store? test space cost of compress of compressed data

sync() operation for blocklists; close() collapses and sync()s ? sync collapses?

BlockSink take ibref initialiser?

cache seek()ed block in readOpen class

inode chunk: flags; [meta] (if flags&0×01); blockref

store priority queue – tuples=pool – multistores

status line queue size / throughput reporting

mount/unpack/ls syntax: / and ./ point at a .vt file, {isotime} for choosing line no later than

leaf level file copy from one store to another

leaf level dir tree copy from one store to another

gdbmstore to use a lock around dbm accesses

gdbmstore – defer file.flush() for speed?

findeEdge – char search – defer until len(data) >= max(len(vocab-word))

use a deque for LRU caches

findWord won’t see words that cross data block boundaries

open/close paradigm for timely shutdown

check stores raise KeyError on no match in getitem

rolling hash better algorithm

findEdge – search on leading char of vocab words

rolling hash:

rolling hash: &0×001fffff, shift left 7, add (ch&0×7f ^ (ch&0×80>>7)), check % 4093 == 1

hash type mapping: 0:sha1, 1:?...

rolling hash: pushable parser for nested data

rolling hash: augument with recognition strings by hash

metadata: inline text in dir, with i: field to indirect to a blockref

don’t compress metadata

dirs: file type: 0=file, 1=dir, 2…

change blockref format: flags(indirect=1,non-sha1hash=2)bs(span)[bs(hashtype)]hash

“vt pull sourceStore”

remote async tagged block interface

decide on new metadata scheme: ACL: u:user:rw-x, g:group:rw-x, o:rw-x, m:mtime, a:atime, c:ctime