1 = Performance costs for some common operations =
3 === Publishing an A-byte immutable file ===
7 notes: An immutable file uploaded using convergent encryption will
8 require an additional I/O pass over the entire source file before
9 the upload process can start, since convergent encryption derives
10 the encryption key in part from the contents of the source file.
12 === Publishing an A-byte mutable file ===
14 cost: O(A) + a large constant for RSA + memory usage.
16 notes: Tahoe-LAFS generates a new RSA keypair for each mutable file that
17 it publishes to a grid. This takes up to 1 or 2 seconds on a
20 Part of the process of encrypting, encoding, and uploading a
21 mutable file to a Tahoe-LAFS grid requires that the entire file
22 be in memory at once. For larger files, this may cause
23 Tahoe-LAFS to have an unacceptably large memory footprint (at
24 least when uploading a mutable file).
26 === Downloading B bytes of an A-byte immutable file ===
28 time/cost until the read is satisfied: variable; up to O(A).
29 cost of the entire operation: O(A) if the file isn't cached.
31 notes: When asked to read an arbitrary range of an immutable file,
32 Tahoe-LAFS will download from the beginning of the file up until
33 it has enough of the file to satisfy the requested read.
34 Depending on where in the file the requested range is, this can
35 mean that the entire file is downloaded before the request is
36 satisfied. Tahoe-LAFS will continue to download the rest of the
37 file even after the request is satisfied, so in any case where the
38 file actually has to downloaded from the grid, reading part of an
39 immutable file will result in downloading all of the immutable
40 file. Ticket #798 is a proposal to change this behavior.
42 Tahoe-LAFS will cache files that are read in this manner for a
43 short while, so subsequent reads of the same file may be served
44 entirely from cache, depending on what part of the file they need
45 to read, what part of the file was read by previous reads, and
46 how much time has elapsed since the last read.
48 === Downloading B bytes of an A-byte mutable file ===
52 notes: As currently implemented, mutable files must be downloaded in
53 their entirety before any part of them can be read. We are
54 exploring fixes for this; see ticket #393 for more information.
56 === Modifying B bytes of an A-byte mutable file ===
60 notes: If you upload a changed version of a mutable file that you
61 earlier put onto your grid with, say, 'tahoe put --mutable',
62 Tahoe-LAFS will replace the old file with the new file on the
63 grid, rather than attempting to modify only those portions of the
64 file that have changed. Modifying a file in this manner is
65 essentially uploading the file over again, except that it re-uses
66 the existing RSA keypair instead of generating a new one.
68 === Adding/Removing B bytes in an A-byte mutable file ===
72 notes: Modifying any part of a mutable file in Tahoe-LAFS requires that
73 the entire file be downloaded, modified, held in memory while it
74 is encrypted and encoded, and then re-uploaded. Note that this
75 sort of modification is mostly used internally for directories,
76 and isn't something that the WUI, CLI, or other interfaces will
77 do -- instead, they will simply overwrite the file to be
78 modified, as described in "Modifying B bytes of an A-byte mutable
81 === Adding an entry to an A-entry directory ===
84 notes: In Tahoe-LAFS, directories are implemented as specialized mutable
85 files. So adding an entry to a directory is essentially adding B
86 (actually, 300-330) bytes somewhere in an existing mutable file.
88 === Listing an A entry directory ===
92 notes: Listing a directory requires that the mutable file storing the
93 directory be downloaded from the grid. So listing an A entry
94 directory requires downloading a (roughly) 330 * A byte mutable
95 file, since each directory entry is about 300-330 bytes in size.
97 === Checking an A-byte file ===
99 cost: variable; between O(N) and O(S), where N is the number of shares
100 generated when the file was initially uploaded, and S is the
101 number of servers on your grid.
103 notes: To check a file, Tahoe-LAFS queries the servers that it knows
104 about until it either runs out of servers, or finds all of the
105 shares that were originally uploaded. Note that neither of these
106 values directly depend on the size of the file. This is
107 relatively inexpensive, compared to the verify and repair
110 === Verifying an A-byte file ===
114 notes: To verify a file, Tahoe-LAFS downloads all of the ciphertext
115 shares that were originally uploaded to the grid and integrity
116 checks them. This is, for well-behaved grids, likely to be more
117 expensive than downloading an A-byte file, since only a fraction
118 of these shares are necessary to recover the file.
120 === Repairing an A-byte file (mutable or immutable) ===
122 cost: variable; up to around O(A)
124 notes: To repair a file, Tahoe-LAFS generates and uploads missing shares
125 in the same way as when it initially uploads the file. So,
126 depending on how many shares are missing, this can be about as
127 expensive as initially uploading the file in the first place.