source: trunk/docs/historical/peer-selection-tahoe3.txt

Last change on this file was aa2c693, checked in by Brian Warner <warner@…>, at 2008-06-03T01:38:32Z

move historical docs from wiki pages into the source tree, clearly marked as historical

  • Property mode set to 100644
File size: 4.1 KB
Line 
1= THIS PAGE DESCRIBES HISTORICAL ARCHITECTURE CHOICES: THE CURRENT CODE DOES NOT WORK AS DESCRIBED HERE. =
2
3When a file is uploaded, the encoded shares are sent to other peers. But to
4which ones? The PeerSelection algorithm is used to make this choice.
5
6In the old (May 2007) version, the verifierid is used to consistently-permute
7the set of all peers (by sorting the peers by HASH(verifierid+peerid)). Each
8file gets a different permutation, which (on average) will evenly distribute
9shares among the grid and avoid hotspots.
10
11This permutation places the peers around a 2^256^-sized ring, like the rim of
12a big clock. The 100-or-so shares are then placed around the same ring (at 0,
131/100*2^256^, 2/100*2^256^, ... 99/100*2^256^). Imagine that we start at 0 with
14an empty basket in hand and proceed clockwise. When we come to a share, we
15pick it up and put it in the basket. When we come to a peer, we ask that peer
16if they will give us a lease for every share in our basket.
17
18The peer will grant us leases for some of those shares and reject others (if
19they are full or almost full). If they reject all our requests, we remove
20them from the ring, because they are full and thus unhelpful. Each share they
21accept is removed from the basket. The remainder stay in the basket as we
22continue walking clockwise.
23
24We keep walking, accumulating shares and distributing them to peers, until
25either we find a home for all shares, or there are no peers left in the ring
26(because they are all full). If we run out of peers before we run out of
27shares, the upload may be considered a failure, depending upon how many
28shares we were able to place. The current parameters try to place 100 shares,
29of which 25 must be retrievable to recover the file, and the peer selection
30algorithm is happy if it was able to place at least 75 shares. These numbers
31are adjustable: 25-out-of-100 means an expansion factor of 4x (every file in
32the grid consumes four times as much space when totalled across all
33StorageServers), but is highly reliable (the actual reliability is a binomial
34distribution function of the expected availability of the individual peers,
35but in general it goes up very quickly with the expansion factor).
36
37If the file has been uploaded before (or if two uploads are happening at the
38same time), a peer might already have shares for the same file we are
39proposing to send to them. In this case, those shares are removed from the
40list and assumed to be available (or will be soon). This reduces the number
41of uploads that must be performed.
42
43When downloading a file, the current release just asks all known peers for
44any shares they might have, chooses the minimal necessary subset, then starts
45downloading and processing those shares. A later release will use the full
46algorithm to reduce the number of queries that must be sent out. This
47algorithm uses the same consistent-hashing permutation as on upload, but
48instead of one walker with one basket, we have 100 walkers (one per share).
49They each proceed clockwise in parallel until they find a peer, and put that
50one on the "A" list: out of all peers, this one is the most likely to be the
51same one to which the share was originally uploaded. The next peer that each
52walker encounters is put on the "B" list, etc.
53
54All the "A" list peers are asked for any shares they might have. If enough of
55them can provide a share, the download phase begins and those shares are
56retrieved and decoded. If not, the "B" list peers are contacted, etc. This
57routine will eventually find all the peers that have shares, and will find
58them quickly if there is significant overlap between the set of peers that
59were present when the file was uploaded and the set of peers that are present
60as it is downloaded (i.e. if the "peerlist stability" is high). Some limits
61may be imposed in large grids to avoid querying a million peers; this
62provides a tradeoff between the work spent to discover that a file is
63unrecoverable and the probability that a retrieval will fail when it could
64have succeeded if we had just tried a little bit harder. The appropriate
65value of this tradeoff will depend upon the size of the grid, and will change
66over time.
Note: See TracBrowser for help on using the repository browser.