Social Clouds provide the capability to share resources among participants within a social network - leveraging on the trust relationships already existing between such participants. In such a system, users are able to trade resources between each other, rather than make use of capability offered at a (centralized) data centre. Incentives for sharing remain an important hurdle to make more effective use of such an environment, which has a significant potential for improving resource utilization and making available additional capacity that remains dormant. We utilize the socio-economic model proposed by Silvio Gesell to demonstrate how a "virtual currency" could be used to incentivise sharing of resources within a "community". We subsequently demonstrate the benefit provided to participants within such a community using a variety of economic (such as overall credits gained) and technical (number of successfully completed transactions) metrics, through simulation.
1. Incentivising Resource Sharing in
Social Clouds
Magdalena Punceva
Joint work with: Ivan Rodero (Rutgers University), Manish Parashar
(Rutgers University), Omer Rana (Cardiff University) and Ioan Petri
(Cardiff University)
2. Social Clouds
• What is a social cloud?
– Resource sharing system built on top of an existing social network.
• Purpose: sharing resources among participants in social
networks
– Resource: storage, computational power, applications
– Social networks: Facebook, LinkedIn, Twitter…
• Why social clouds?
– Utilize trust relationships already existing between such participants
(in contrast to p2p sharing).
3. Social Clouds
3-hops
sharing
2-hops
sharing
1-hop
sharing
4. Trust in social networks
• Trust
– Inherent in social networks: friends trust each other
– People will be more willing to share resources with friends or socially
close users then with total strangers.
– Trust levels are variable
– By trust we mean trusting that a friend will not misbehave (e.g. a
friend will not interrupt a resource exchange or transaction).
5. Incentives and Trading
• Incentives
– Although trust exists between friends, incentives are needed to
motivate the users to share their spare resources.
– Incentives remain an important hurdle to make effective use of social
clouds environment.
The central problem in this work is defining the right incentives
for sharing in social clouds.
• Trading as an incentive
- Users will trade resources between each other and get
payed for the resources they share.
6. Problem statement
• Key challenges: propose economic incentives for sharing
resources that satisfy the following goals:
1. Node-providers who offer good quality services and resources
should get an advantage
2. Node-consumers should be able to report and share their
experiences and the feedback should affect the payoff providers
receive.
3. Distributed solution
• Existing related approaches
– Barter: simple however limiting [BitTorrent}
– Credit-networks: p2p sharing [Z. Liu et al., P. Dandekar et al.]
– Global currency: complex rules [C. Aperjis et al.,V. Vishnamurthy et al. ,
B. Yang et al.]
7. Related approach: credit networks
u v w
c1 c2
• Node u trusts node v for up to c1 units of v’s
currency (v can use a service from u for up to c1
units of v’s currency)
• All nodes use the same currency
• Nodes participate in an underlying social network.
• Credit limits c1 and c2 reflect the level of trust
between u and v and v and w respectively.
8. Related approach: credit networks
• Transaction: w purchases a product/service from u
worth p units.
p p
u v w
• Transaction goes through a chain of friends (1-hop
neighbors).
9. Related approach: credit networks
• After transaction: credit limits are being decreased
on each link p units.
p p
u v w
c1-p c2-p
In order for a transaction to be successful: c1>p and
c2>p
There must be at least p credits on every link
10. Our approach: distributed currency
• Each node generates its own currency.
• Currency values may be different.
• Trading is done using such virtual currencies.
• When a node pays to another node, currency exchange rates
must be known to both.
• Partially inspired by Silvio Gesell’s work: The Natural Economic
Order, 1958.
Idea: The value of a node’s currency depends on the quality of
the resources/services it offers.
11. How to define currency exchange
rates?
• Currency exchange rates should satisfy the following conditions:
1) Common knowledge: nodes should know the exchange rates
2) Conservation: currency exchange rates should be conserved along any cycle of
payment.
A
1 B-dollar=2 A-dollars
1 C-dollar=3 B-dollars
1/2 ? Exchange rate between A
and C must be 1/6 in
order to conserve the
currencies.
B C
1/3
12. Clusters of trust
• The requirements (common knowledge and conservation) imply globally
defined exchange rates. Is distributed model possible?
• Our solution: clusters of trusted (socially close) nodes.
• Currency exchange rates are defined within each cluster.
13. How to define the exchange rates
within a cluster?
• Value of a node’s currency depends on the quality of its
resources.
• Consumers give feedback as a score about the providers ->
reputation model
• The reputation of a node is an average of all received scores .
14. Two types of payments
Transaction 2
Transaction 1
• Two types of payments: within a cluster (Transaction 1) and between
clusters (Transaction 2).
15. Payments within a cluster
(Transaction 1)
• Reputation lists are maintained within each cluster: e.g. a
list (r1,r2,..,rn) corresponds to nodes 1…n that belong to
cluster 1
• Reputation scores are given upon successful transaction.
• Reputation of a node is an average of all scores received.
• Currency conversion rates:
1u’s dollar=(ru/rv) v’s dollars
u v
ru rv
16. Payments between clusters
(Transaction 2)
• As in credit networks: nodes exchange IOU (I owe you) credits.
• Such credits have limited use: if node u has p IOUs from node
v, then can use them to purchase service/product only from v.
• Convertible currencies can be used for purchasing
services/products from any node in the corresponding cluster.
• Simple to implement, supports asynchronous demands,
simpler than price forming mechanisms.
17. Our solution: summary
Nodes who offer good Their currencies will have
services should get an higher values since they
advantage depend on reputations
Consumers should be able Reputation model:
to give feedback and share aggregates feedback
their experiences scores
Clusters provide
Distributed solution decentralized and self-
organized solution
18. Experimental setup
• Java based simulator
• Synthetic social graph based on measurements study about
Microsoft IM communication graph
– p(k)≈k-ae-bk
– av. clustering coefficient: 0.37
• Main metric: number of successful transactions, account
statements
• Experiment: set of predefined transactions
• Transaction path: shortest path (Dijkstra algorithm).
19. Our simulations should answer
these questions
• How does the number and sizes of clusters affect the number
of transactions completed? How much do we gain in terms of
completed transactions compared to pure credit networks?
• Is the approach scalable?
• How much does the non-uniform (power-law) distribution of
reputations and social graph degrees affect the successfully
completed transactions?
20. Our results: impact of cluster sizes
Success rate increases non-linearly with cluster sizes.
21. Our results: impact of reputation
distribution
Equal and uniform reputation distributions lead to higher
success rate than the power-law distribution.
22. Our results: scalability and impact of
the social graph
n 256 512 1024 2048 4096
success 76.00 72.86 81.70 78.70 70.65
rate(%)
23. Our results: impact of account limits
Number of successful transaction almost linearly
increases with credit limits.
24. Conclusions
• We extended the credit-network approach by enabling
within clusters currency conversions.
• By simulations we have shown how much it improved
long-term liquidity (achieved higher number of
completed transactions).
• Different currency values give advantage to high-
quality providers (incentive for improving the quality of
resources)
• Distributed reputation model
• Scalable with social graphs’ sizes and structures.
25. Research directions
• Integrating with CometCloud: parallelization and
application.
• Solutions for non-cooperative nodes: nodes may
downrate good providers, make coalitions to increase
reputation mutually.
• Free money property: money loses value over time.
• Exchange rate should include the impact of demand
and predefined quality of service.
• Network dynamics: nodes join/leave the network.
• Cluster dynamics: nodes join/leave a cluster.