Professional Documents
Culture Documents
Impact of Neighbor Selection On Performance and Resilience of Structured P2P Networks
Impact of Neighbor Selection On Performance and Resilience of Structured P2P Networks
Abstract 1
Cumulative probability
mance of peer-to-peer overlay networks. While its impact
0.6
on performance has been recognized, few have examined
the impact of neighbor selection on network resilience. In
0.4
this paper, we study the impact with a generalized cost
model for overlay construction that takes into considera-
tion different types of heterogeneity, such as node capac- 0.2
3 Structured Overlay Construction where a is the neighbor index, O W is the neighbor indexed
by a , W is the number of neighbors, b W is the set of desti-
In the construction of structured peer-to-peer networks, nations routed through the neighbor O W , Ac@!\D is the node
each node chooses neighbors that meet logical identifier cost value of , @%dB3e%D is the edge cost between two nodes
constraints (e.g., prefix matching or identifier range), and d and e , and @)N#D is the edge cost of N .
builds directional links to them. These constraints are Depending on the optimization goal, we can choose dif-
flexible such that a number of nodes are possible neigh- ferent metrics for and , including latency, through-
bors for each routing table entry. Intelligent selection of put, reliability, availability, monetary cost, or any com-
neighbors from the set of possible neighbor nodes signif- bination thereof. For example, choosing high capacity
icantly impacts the overlay’s performance, resilience, and nodes as neighbors can decrease lookup latency and in-
load balancing properties. crease the overall lookup processing capacity of the sys-
The neighbor selection problem can be reduced to a tem. On the other hand, using availability as a metric cre-
generalized cost minimization problem. We present here ates a more stable network.
a generalized cost model that captures general node and
link characteristics during neighbor selection. Ideally, op- Note that our idealized cost function assumes full
timizing neighbor selection for node means minimizing knowledge of the network components, and is therefore
the sum of the cost from to all other nodes. The cost not feasible in practice. Since most peer-to-peer proto-
from to consists of two factors: cost incurred by inter- cols focus on optimizing neighbor tables locally, we will
mediate overlay nodes (node cost: ) and cost incurred focus on the application of our cost function to the cost
by overlay network links (edge cost: ). Let be the of the first overlay hop. Therefore, we focus on neighbor
network size. The cost of node (
) is: selections that consider the first hop and optimize latency
under uniform traffic (?A@<CB%EDgfihBjBI ).
!#"%$'&(!#" where
*,+.-0/ 1 * 983+.:;/ <1 8 (1)
$ & )#" 32 $ 45"76 32 $ < =>" Table 1 shows the four neighbor selection cost func-
tions. Random chooses neighbors randomly. Dist chooses
where ?A@!CB%ED is the traffic from to , GFH@!BIED is the cost neighbors physically closest in the network to adapt to the
of the path from to , JK@<CB%ED is the path (a set of edges) underlying network topology. Currently, Bamboo, Pastry,
from to , LM@!BIED is the set of intermediate overlay nodes and Tapestry use this mechanism. Cap chooses neighbors
in the path JK@<CB%ED (it does not include and ), N is the that have the smallest processing delay. CapDist chooses
edge in the path JK@<CBI(D , O is the node in LP@<CB%ED , @!O0D neighbors that gives the smallest combined latency, which
is the cost of node O , and .@!N#D is the cost of edge N . If is the sum of the node processing delay and the overlay
?A@<CBI(D =0, there is no incentive for the node to optimize link delay.
4.5 1
3.5 0.8
Cumulative probability
3
Lookup latency
0.6
2.5
2
0.4
1.5
Figure 2: Average lookup latency for uniform processing delay Figure 4: CDF of the number of incoming edges for uniform
distribution processing delay distribution
6
3 overlay node placements on each topology.
5
For our Chord experiments, each node forwards mes-
4
sages to the live neighbor that is closest to the destination
Lookup latency
selection algorithms. We also examine the impact such is the maximum processing delay. Figure 2 shows aver-
algorithms have on the static resilience of the resulting age lookup latency over all node pairs in Tapestry. By
overlay to randomized failures and targeted attacks. exploiting network proximity and heterogeneous capac-
ity, CapDist achieves the best lookup performance. When
processing delay variation is high (
=1s), CapDist per-
4.1 Simulation Setup
forms 30 better than Dist and 48 better than Ran-
We simulate the Tapestry [18] and Chord [13] protocols dom. When no variation exists, (i.e.,
=0s), Dist and
as representatives of their respective geometries (tree and CapDist exploit network proximity to outperform Ran-
ring). When each node optimizes its cost function, it per- dom and Cap.
forms random sampling to select neighbors and choose We now look at a bimodal model for processing capac-
the best one among the samples. In our experiments, we ity, where nodes are either fast or slow. Fast nodes pro-
use 32 samples for each routing level in Tapestry or each cess 100 lookup messages per second while slow nodes
finger in Chord. process 1 message per second. Figure 3 shows
that as we
Our simulations use 5100 node transit-stub network vary the fraction of fast nodes from 0 to 20 , neighbor
topologies generated using the GT-ITM library [15]. We selection using capacity (Cap and CapDist) favors routes
construct Chord and Tapestry overlays of 4096 nodes through fast nodes and achieves better performance. For
by placing overlay nodes to random physical locations. instances where the variation in processing capacity is ex-
We gather results with 9 different configurations for GT- tremely high, we expect that capacity utilization at fast
ITM, generate 3 transit-stub topologies each, and choose nodes will be limited by the routing constraints of the pro-
1 1 1
1 1
0.8 0.8
Failed paths (%)
0.4 0.4
(a) (b)
Figure 6: Chord under random node failures. (a) Chord varying finger selection on finger table, (b) Chord varying finger selection
on finger table and having 4 sequential neighbors.
tocol, and the deployment of virtual nodes is necessary to live endpoints that can still route to each other via the
fully exploit the excess processing capacity. overlay after an external event, either randomized node
Using latency optimization creates uneven distributions failures or targeted attacks. We assume attacks focus on
of nodes’ incoming node degrees. Nodes near the cen- removing nodes with the highest in-degree in order to
ter of the network (i.e., transit domains) and nodes with maximize damage to overall network reachability. For
high capacity are preferred, and minimize path latency by these experiments, we assume nodes have an uniform pro-
utilizing low latency links or low processing delay. Fig- cessing delay distribution with
= 0.5s.
ure 4 shows the cumulative distribution function (CDF) For Tapestry, we examine resilience of the base proto-
of nodes’ in-degrees in Tapestry networks with different col, the base protocol plus additional backup routes (all
neighbor selection algorithms. Unlike Random, results chosen using a number of neighbor selection algorithms),
from cost-optimized overlays show slow transitions and and the base protocol plus backup routes chosen at ran-
long tails. We also observe that the CDF of nodes in tran- dom. Note that adding two backup links triples the num-
sit domains is more skewed and has longer tails than that ber of neighbors. For Chord, we examine the base proto-
of nodes in stub domains. col and the base protocol plus sequential neighbors.
1 1
0.8 0.8
Failed paths (%)
0.4 0.4
(a) (b)
Figure 8: Chord nodes under attack. (a) Chord varying finger selection on finger table, (b) Chord varying finger selection on finger
table and having 4 sequential neighbors.
tively low impact on overall connectivity. work such that powerful nodes have significantly higher
Figures 5 and 6 show the failure tolerance of Tapestry in-degrees than weaker nodes. This means that like the
and Chord, respectively. Surprisingly, we see the fail- unstructured networks studied by Albert et al., structured
ure tolerance is a little affected by neighbor selections. peer to peer overlays that consider capacity in neighbor
The tighter outgoing link constraints of structured peer to selection are vulnerable to attacks.
peer networks allow less variation in the resulting topol- As shown in Figures 7 and 8, attacking nodes with high
ogy than unstructured networks. Every node has at least in-degree affects network connectivity severely. Random
@!eE D outgoing links, and randomized naming also shows the best attack tolerance among neighbor selec-
smoothens out distribution of outgoing edges. Since each tions. CapDist has worse attack tolerance than Dist, al-
lookup takes @)eH D hops regardless of neighbor selec- though it has better performance as shown in Section 4.2.
tion cost functions, the probability of meeting randomly This result demonstrates a fundamental tradeoff be-
failed nodes in a lookup will be similar. tween performance and attack resilience in structured
Adding backup links in Tapestry and sequential neigh- overlay construction. The performance gain from neigh-
bors in Chord dramatically improves failure tolerance. bor selection algorithms increases the variability of in-
Note that in Tapestry, failure curves change from an ex- degrees among nodes. Nodes with high capacity or nodes
ponentially increasing curve to a smooth S-shaped curve near the center of the network end up with high in-degrees
due to path redundancy when backup links are added. and have a disproportionately large impact on network
connectivity when they fail.
4.3.2 Targeted Node Attacks
4.3.3 Analysis of Extra Redundancy
While structured peer to peer protocols define a minimum
number of outgoing links per node, a node’s number of in- We observe that adding backup links or sequential neigh-
coming links is unrestricted. This means that neighbor se- bors can increase attack tolerance significantly. Ran-
lection algorithms considering capacity will skew the net- domly choosing backup links in Tapestry and sequential
neighbors in Chord avoids routing hotspots that are vul- [3] C HAWATHE , Y., R ATNASAMY, S., B RESLAU , L., L AN -
nerable to targeted attacks. In Tapestry for example, cost- HAM , N., AND S HENKER , S. Making gnutella-like p2p
optimized backup links are less effective at improving at- systems scalable. In Proc. of ACM SIGCOMM (2003).
tack tolerance than random backup links. Paying the ad- [4] C HUN , B., C ULLER , D., ROSCOE , T., BAVIER , A., P E -
ditional cost of maintaining extra links improves static re- TERSON , L., WAWRZONIAK , M., AND B OWMAN , M.
silience against targeted attacks. Planetlab: An overlay testbed for broad-coverage services.
Another technique to improve tolerance is to impose In ACM Computer Communication Review (July 2003).
a maximum in-degree per node, but this restriction in- [5] G UMMADI , K. P., G UMMADI , R., G RIBBLE , S. D., R AT-
creases lookup latency. Bounding the degree of over- NASAMY, S., S HENKER , S., AND S TOICA , I. The impact
lay nodes can be used as a defense against Eclipse at- of dht routing geometry on resilience and proximity. In
tacks [12]. Proc. of ACM SIGCOMM (2003).
[6] R AO , A., L AKSHMINARAYANAN , K., S URANA , S.,
K ARP, R., AND S TOICA , I. Load balancing in structured
p2p systems. In Proc. of IPTPS (2003).
5 Conclusion [7] R ATNASAMY, S., F RANCIS , P., H ANDLEY, M., K ARP,
R., AND S HENKER , S. A scalable content-addressable
Previous research argued for the consideration of network network. In Proc. of SIGCOMM (August 2001), ACM.
or physical characteristics of nodes in overlay construc-
[8] R ATNASAMY, S., H ANDLEY, M., K ARP, R., AND
tion. In this paper, we take a quantitative approach to ex-
S HENKER , S. Topologically-aware overlay construction
amining the benefits and costs of considering such criteria and server selection. In Proc. of IEEE INFOCOM (2002).
in overlay construction.
[9] R HEA , S., G EELS , D., ROSCOE , T., AND K UBIATOW-
We present a generalized model for neighbor selection
ICZ , J. Handling churn in a dht. In Proc. of the USENIX
that incorporates metrics for network proximity and avail- Annual Technical Conference (June 2004).
able resources (capacity), and show that while consid-
ering these factors can lead to significant gains in rout- [10] ROWSTRON , A., AND D RUSCHEL , P. Pastry: Scalable,
distributed object location and routing for large-scale peer-
ing performance, these benefits come with their associ-
to-peer systems. In Proc. of Middleware (Nov 2001),
ated costs. We find that the choice of neighbor selection ACM.
algorithm drives a tradeoff between performance and re-
silience to attacks. [11] S AROIU , S., G UMMADI , P. K., AND G RIBBLE , S. D. A
measurement study of peer-to-peer file sharing systems. In
Optimized structured overlays have unbalanced struc- Proc. of MMCN (2002).
tures. These overlays do not bound the number of in-
[12] S INGH , A., C ASTRO , M., D RUSCHEL , P., AND ROW-
coming links per node. Thus central nodes in a network
STRON , A. Defending against eclipse attacks on overlay
or nodes with more resources will have much higher in-
networks. In Proc. of the ACM SIGOPS European Work-
degree than others. Should high degree nodes be attacked, shop (September 2004).
the impact on network connectivity is severe. On the other
[13] S TOICA , I., M ORRIS , R., K ARGER , D., K AASHOEK ,
hand, the minimum out-degree means even for overlays
M. F., AND BALAKRISHNAN , H. Chord: A scalable peer-
that optimize towards proximity or available resources,
to-peer lookup service for internet applications. In Proc.
most nodes achieve enough resilience against randomized of SIGCOMM (August 2001), ACM.
failures.
[14] WALDVOGEL , M., AND R INALDI , R. Efficient topology-
As future work, we intent to investigate the resilience aware overlay network. In Proc. of HotNets (2002).
of different geometries under different neighbor selection
algorithms. We also plan to investigate the impact of these [15] Z EGURA , E. W., C ALVERT, K. L., AND B HATTACHAR -
JEE , S. How to model an internetwork. In Proc. of IEEE
neighbor selection algorithms on dynamic resilience, such
INFOCOM (1996).
as when maintenance algorithms repair failures over time.
[16] Z HANG , H., G OEL , A., AND G OVINDAN , R. Incremen-
tally improving lookup latency in distributed hash table
systems. In Proc. of ACM SIGMETRICS (2003).
References [17] Z HAO , B. Y., D UAN , Y., H UANG , L., J OSEPH , A., AND
K UBIATOWICZ , J. Brocade: Landmark routing on overlay
[1] A LBERT, R., J EONG , H., AND BARABASI , A.-L. Er-
networks. In Proc. of IPTPS (2002).
ror and attack tolerance of complex networks. Nature 406
(July 2000), 378–381. [18] Z HAO , B. Y., H UANG , L., R HEA , S. C., S TRIBLING , J.,
J OSEPH , A. D., AND K UBIATOWICZ , J. D. Tapestry: A
[2] C ASTRO , M., D RUSCHEL , P., H U , Y. C., AND ROW- global-scale overlay for rapid service deployment. IEEE
STRON , A. Exploiting network proximity in peer-to-peer J-SAC 22, 1 (January 2004), 41–53.
overlay networks, technical report msr-tr-2002-82, 2002.