AbstractโMany decentralized online social networks (DOSNs)
have been proposed due to an increase in awareness related to
privacy and scalability issues in centralized social networks. Such
decentralized networks transfer processing and storage functionalities
from the service providers towards the end users. DOSNs
require individualistic implementation for services, (i.e., search,
information dissemination, storage, and publish/subscribe). However,
many of these services mostly perform social queries, where
OSN users are interested in accessing information of their friends.
In our work, we design a socially-aware distributed hash table
(DHTs) for efficient implementation of DOSNs. In particular,
we propose a gossip-based algorithm to place users in a DHT,
while maximizing the social awareness among them. Through
a set of experiments, we show that our approach reduces the
lookup latency by almost 30% and improves the reliability of
the communication by nearly 10% via trusted contacts.
Beginners Guide to TikTok for Search - Rachel Pearson - We are Tilt __ Bright...
ย
Socially-Aware Distributed Hash Tables for Decentralized Online Social Networks
1. Socially-Aware Distributed Hash Tables for
Decentralized Online Social Networks
Muhammad Anis Uddin Nasir, Sarunas Girdzijauskas, Nicolas Kourtellis
KTH Royal Institute of Technology
Telefonica Research
Socially-aware DHTs for DOSNs 1
7. Research Question
โCan we design an overlay for decentralized
online social networks that can support light
weight, low-cost devices such as mobile
phones, web browsers"
Socially-aware DHTs for DOSNs 7
8. Decentralized Online Social Networks
โข Low Management Overhead
โข Scalability
โข Bounded Degree
โข Reliability and Security
Socially-aware DHTs for DOSNs 8
13. Problem Formulation
โข
โ where V is the set of vertices and E is the set of edges,
connecting the vertices.
โข A
โ Neighbors of node i are all the nodes that share an edge
with node i
Socially-aware DHTs for DOSNs 13
17. Neighbor Selection
โข Direct: A node selects one of his friends in the
social graph uniformly at random.
โข Greedy: A node selects its friend with
strongest tie.
โข Smart: A node selects a node m uniformly at
random from its top k strongest friends.
โข Random: A node i selects a random other
node.
Socially-aware DHTs for DOSNs 17
21. Experiments
โข What is the tuning cost of the algorithm?
โข How does node ordering impact algorithm
convergence?
โข What are the performance gains with respect
to lookup latency and reliability?
Socially-aware DHTs for DOSNs 21
24. Q1: Peer Selection Scheme
โข Direct, Random and Smart Neighbor selection
schemes perform similar
Socially-aware DHTs for DOSNs
3
3.5
4
4.5
5
0 100 200 300 400 500 600 700 800 900 1000
LookupLatency
Number of iterations
Performance
Random Neighbor
Direct Neighbor
Greedy Neighbor
Smart Neighbor
24
25. Q2: Execution Ordering
Socially-aware DHTs for DOSNs
3
3.5
4
4.5
5
50 100 150 200 250 300 350 400 450 500
LookupLatency
Number of iterations
Performance
Descending Order
Direct Peer Sampling
Ascending Order
โข Convergence is fast with ordered execution of the
algorithm
25
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
50 100 150 200 250 300 350 400 450 500
FractionofMigrationCost
Number of Iterations
Fraction of Nodes Swapping Identifiers
Descending Order
Direct Peer Sampling
Ascending Order
26. Q3: Performance
โข SD has less improvement due to lack of
clustering in the social graph
3
3.5
4
4.5
5
5.5
6
6.5
7
FB W
V
SD TW
LookupLatency
Symphony
Direct
Socially-aware DHTs for DOSNs 26
27. Q3: Reliability
โข Reliability improves significantly in terms of finger
table
0.01
0.1
1
10
FB W
V
SD TW
Reliability1%
Symphony
Direct
Socially-aware DHTs for DOSNs 27
28. Q3: Reliability
โข Significant improvement in the reliability in terms of
connections
0.1
1
10
100
FB W
V
SD TW
Reliability2%
1-hop
2-hop
3-hop
Socially-aware DHTs for DOSNs 28
29. Conclusion
โข Socially-aware distributed hash tables improves the
performance for decentralized online social network
โข We propose a gossip-based algorithm that considers
social ties to achieve social-awareness in DHT
โข We show that our approach reduces the lookup
latency by almost 30% and improves the reliability of
the communication by nearly 10% via trusted
contacts.
Socially-aware DHTs for DOSNs 29
30. Socially-Aware Distributed Hash Tables for
Decentralized Online Social Networks
Muhammad Anis Uddin Nasir, Sarunas Girdzijauskas, Nicolas Kourtellis
KTH Royal Institute of Technology
Telefonica Research
Socially-aware DHTs for DOSNs 30
32. Dealing with Failures
โข Social graphs evolve in a linear fashion
โข Mobility or inactivity of peers, and peer or
network failure cause instability
Socially-aware DHTs for DOSNs 32
33. Q1: Migration Cost
Socially-aware DHTs for DOSNs
0.001
0.01
0.1
1
0 100 200 300 400 500 600 700 800 900 1000
FractionofMigrationCost
Number of Iterations
Migration Cost
Minimizing Hop Distance
Minimizing Euclidean Distance
33
Editor's Notes
Small world network
Requirement
DHT
Random DHT high latency and less reliability
Both network has small world property,
We should be able to embed
We consider social ties
Gossip based algorithm
Peer selection schemes
Experimental section
High Clusterization
Low Diameter
The incentive for a provider is the access to large amounts of data, which can be used for business-related purposes
However, these incentives have raised privacy concerns among users.
Each device only supports limited number of connections
DHTs are small-world graphs embedded in some identifier space, where we know how navigate efficiently. Then you discuss the properties of social networks, and that they also are small worlds. Then you say what we would like to take a subgraph of social-netowrk, such that it resembles the properties of a DHT graph and intelligently embed it into some id space. if we do so, we can achieve navigability as in classical DHT, but mostly using only social links, with all the good "consequences": fast news-feed creation, low latency, low "spam/relay" nodes etc.
DHTs are a very promising solution for DOSNs since they provide all required functionalities with a limited peer degree in the resulting overlays.
Such overlays are small-world in nature and have efficient routing properties.
However, current DHTs create such overlays solely based on the peer IDs which are assigned uniformly at random and do not reflect the social graph structure of DOSNs.
This significantly downgrades the performance of DHT-based DOSNs since most of the workloads directly reflect the topology of the social graph mapped on the overlay.
On a DHT- based DOSN, such requests would correspond to generating expensive relay traffic for performing simple actions on each social link.
Nodes get an identifier from an identifier space
Each node is map to a ring, where each node points to two direct neighbors (predeccessor and succeessor )
Each node further creates a long range link using a probability distribution function
That guarantees small world nature by sampling more short range links and few long range links
Lets take a toy example to understand the problem
We use a ring based DHT. And place users randomly in the form of the ring
Each user wants to communicate with his friends, which are shown in the example with the same color
Now when a user create a request
In a socially aware DHT we want to map each user close to their friends for example this
The benefits of this approach is two fold
First, routing requires less number of steps
Second, routing involves only friend nodes that improves the overall reliability of the system as social friends are more likely to forward your message
Third, overlay becomes simple and easy to maintain
For a DHT, we define the distance between two nodes i and j as dij .
We define two different distance metrics: 1) euclidean distance between the two node ids, and 2) lookup latency (number of hops in the overlay) between the two nodes.
we aim to improve the performance of DHT- based DOSN services by designing a socially-aware DHT
One the node is selected, the node will try to move closer to the node.
How can we do it ?
One easy way of doing it is to swap identifier to any of the neighbors of node A
In a socially aware DHT we want to map each user close to their friends for example this
The benefits of this approach is two fold
First, routing requires less number of steps
Second, routing involves only friend nodes that improves the overall reliability of the system as social friends are more likely to forward your message
Third, overlay becomes simple and easy to maintain
We run algorithm for 1000 iterations
With four different peer selection schemes
On y-axis we report the lookup latency
Results show that the greedy node selection has the fastest convergence time. However, it provides minimal improvements with respect to lookup latency.
The other three approaches, i.e., random, direct and smart selection, achieve better improvements in the performance of the overlay.
Therefore, we use direct peer selection in the further experiments due to its simple and decentralized nature.
On the one hand, top degree nodes are more important in the network and can affect many nodes in the overlay at once, thus, affecting its speed of convergence (could be faster, or may lead to oscillations).
On the other hand, bottom degree nodes are more periphery nodes and may allow slower, but steadier, convergence.
Small world network
Requirement
DHT
Random DHT high latency and less reliability
Both network has small world property,
We should be able to embed
We consider social ties
Gossip based algorithm
Peer selection schemes
Experimental section
Gossip-based algorithm ensures handling evolving graphs
data replication
providing guarantees for eventual consistency
socio-incentivized networks
We compare the migration cost of two different overlay distance function