6.5
0.9
3
10
1000
n u m b e r o f n o d e w ith d e g r e e ( k )
6
number of nodes with degree (k)
2
0.8
10
800 5.5
number of hops
1 0.7
10 5
600
0 0.6 4.5
10
0 1 2
10 10 10
degree (k)
4
400
0.5
3.5
0.3 2.5
0 2000 3000 4000 5000 6000 7000 8000 9000 10000 0 2000 4000 6000 8000 10000 12000 14000 16000 18000
0 20 40 60 80 100 120
node life time (sec) number of nodes
degree (k)
2.2 Our proposal larger knowledge of the network. These nodes, when used
Additional data structure: In S-Chord, a node for routing, would accelerate the query search process.
maintains a new table, that we call a finger selector table 2.3 Simulation results
(FST) in addition to the classic finger table. If there is a We compare the performance of our proposed search
finger link from one node to its finger, we add a reverse link model S-Chord to the classic Chord DHT. We simulated a
that links the same two nodes, but in the reverse direction. network with N = 2k nodes. We varied k from 9 to 14.
These reverse links connect the node to its finger selectors Figure 4 (c) shows that S-Chord exhibits an improved
(FS), i.e. the nodes that selected it as a finger. In addition to behavior compared to standard Chord in term of average
the finger selector ID information, the FST stores the search path length. The facts that the query traverses highly
predecessor ID of the corresponding node. connected nodes, that every node keeps a sort of index of
Knowing the FS predecessor ID, every node may know the its neighbors shared content, and that routing is performed
data interval associated to each FS entry, which in both directions, accelerate considerably the query
corresponds to the index of the content of the process. Figure 4 (b) plots the request success rate (RSR)
corresponding node, and that without exchanging content as a function of nodes mean life time. We note that S-
information with neighbors, and consequently without Chord outperforms Chord and especially for shorter life
generating extra overhead. The interesting property of this time values (over 73% for all values). For such short
addition is that it comes at almost no additional cost. lifetimes, in Chord, the success ratio decreases to less than
Routing: Structured DHT networks perform key based 60%.
routing. In fact, the query follows a calculated path
according to the encountered routing tables and the 3. CONCLUSION
searched key, regardless of the nodes connectivity. In this paper, we propose a new super-peer based lookup
In order to shorten the query search process in Chord, we algorithm without imposing a hierarchical architecture.
propose to use the finger selector table in addition to the S-Chord employs KBR routing and provides a power law
finger table for the query routing. Therefore, when like broadcast mechanism, to perform query search.
receiving a query, the intermediate node examines first its Primary simulation results show improvements over classic
FST: the node compares the searched key to the IDs of each Chord. The proposed model might be considered as an
finger selector entry and the corresponding predecessor. For alternative to artificially imposing a hierarchical structure.
a given FST entry, if the searched key lies between these As a future work, it would be interesting to apply our model
two IDs, the query is transmitted directly to this finger to different DHTs and evaluate the performance results. For
selector, as it is the owner of the searched key. Thus, the this purpose, we have to adapt our scheme to flexible DHT
lookup search is resolved. If not, the node forwards the geometries such as tree for Pastry or butterfly for Viceroy.
query to its closest preceding finger or finger selector. The 4. REFERENCES
search is performed in the entire neighbourhood. [1] Yuhong Liu, Chris Gauthierdickey, Scalable supernode
S-Chord adds bidirectional links to enlarge the selection in peer-to-peer overlay networks. 2nd
neighborhood knowledge especially for topologically International Workshop on Hot Topics in P2P Systems.
important nodes. Knowing the predecessor ID information, [2] Diestel, Reinhard (2005), Graph Theory (3rd ed.), Berlin,
New York: Springer-Verlag, ISBN 978-3-540-26183-4.
the search is performed in the entire neighbourhood as a
[3] I. Stoica, R. Morris, Chord: a scalable peer-to-peer lookup
local broadcast, but that without flooding the neighbors or service for internet applications, Proc.ACM SIGCOMM
exchanging content information. Therefore, de facto highly 2001 Technical Conf., San Diego, USA, August 2001.
connected Chord nodes (i.e. nodes with a high indegree) are [4] Barabsi, Albert-Lszl "Scale-Free Networks". Scientific
automatically considered super nodes because they use a American, 288:60-69, May 2003.