P2P Applications - Peer-to-Peer Computing: Applications, Architecture, Protocols, and Challenges - page 23

m linearly independent combinations of blocks, can reconstruct the original
chunk using Gaussian elimination.
As we have seen earlier, currently different video streaming systems employ
slightly different methods in “matching” a new peer with a subset of existing
peers watching the same channel. It is, however, not di cult to envision that
a probably more effective approach is to consider the “peer selection” step
as an optimization problem—to minimize the download time, for example.
Indeed, it would be ideal, from a holistic perspective, if such a distributed
peer selection mechanism can achieve:
•Minimum download time for each peer;
•Minimum upload bandwidth required for each peer;
•Maximum content availability; and
•Maximum robustness.
Unfortunately, it is obvious that the above goals are conflicting among each
other. Specifically, while the first two goals are “user-oriented,” the latter two
are “system-oriented.” Putting this in a more incentive-related context, we
can say that the first two goals are individualistic while the latter two concern
social welfare.
It is highly likely that, therefore, any peer selection scheme would not be
able to provide optimal results for all of the above goals. Thus, a more practical
research question is whether we can come up with some e cient distributed
heuristics to achieve near-optimal results for all or some of the above goals.
Bonald et al. [Bonald et al., 2008] consider a “push-based” peer selection
problem. Specifically, in a push-based approach, it is the chunk sender who
selects requesting peers as receivers of its chunks. Let us denote the sending
peer as s and a receiving peer as r. Furthermore, let C(s) and C(r) denote
the set of chunks currently owned by s and r, respectively. Several selection
heuristics are considered:
Random Peer. A receiving peer is randomly chosen from among the list of
currently connected peers.
Random Useful Peer. A receiving peer is randomly chosen from those con-
nected peers r where C(s)/ C(r) = φ. Thus, this is different from the
Random Peer heuristic in that a receiving peer is one which really needs
some chunks from the sender.
Most Deprived Peer. A receiving peer is randomly chosen from those con-
nected peers r where|C(s)/ C(r)|is the largest. The rationale of this
heuristic is to give a higher priority to a peer that needs the most chunks
from the sender.