SuperNova: Super-peers Based Architecture for Decentralized Online Social Networks

1 Introduction

• DOSNs may appeal to privacy-conscious individuals and organizations wary of placing corporate data on public social networks.The paper notes that DOSNs do not automatically ensure scalability, robustness, privacy, or security.Source: §1 · pp. 2–3
• Replication can improve data availability, but choosing storage nodes is difficult under resource constraints and individual behavior; decentralized enforcement is likely infeasible, and reciprocity can leave other nodes with poor availability.Highly available nodes may pair with one another, limiting benefits to the rest of the system.Source: §1 · p. 3
• Super-peers can offer incentives through monetization, recognition, or influence, leverage well-provisioned nodes, and help newcomers who lack established social connections.They may run on end-user computers or cloud services; corporate entities may also run them while retaining control of their data.Source: §1 · p. 3
• SuperNova is a volunteer-agent, super-peer DOSN designed to self-organize by leveraging heterogeneity that may arise from altruism or incentive mechanisms.The design draws on experience with PeerSoN and other decentralized systems, as well as theoretical work.Source: §1 · p. 4
• The study focuses on algorithmic and systems design and evaluation; it defers analysis of changing peer reputation and roles, and excludes security issues essential to practical deployment.The paper evaluates system performance through simulations.Source: §1 · p. 4
• The paper proposes mechanisms for using contributed resources across different user strategies, including where users store data and whose data they store.The system is evaluated for fixed snapshots of strategy mixes, capacities, and churn to assess sustainability and quality of service.Source: §1 · p. 4

2 Related works

• Social-relation-agnostic storage can lack user incentives, while approaches based only on immediate social ties or reciprocity limit benefits to individual neighborhoods and can yield poor availability.PeerSoN uses a DHT; SoJa combines global and socially local storage; SafeBook centers on trust and Matryoshkas; Vis-a-Vis hosts individual user data on cloud-based virtual servers.Source: §2 · p. 5
• SuperNova allows users to store data at friends, super-peers, or strangers and to store others’ data, while super-peers coordinate matching and accounting.Users may pursue reputation and influence toward super-peer status; super-peers may run on commercial clouds or dedicated personal computers.Source: §2 · pp. 5–6
• SuperNova combines bilateral and community-wide incentives, whereas using user resources to distribute files for centralized OSNs leaves their privacy and autonomy issues in place.Source: §2 · pp. 5–6
• SuperNova explicitly provides for new nodes to join using super-peer and stranger resources while they establish or discover adequate friendships.Source: §2 · p. 6
• The paper evaluates parameter choices and user-strategy mixes experimentally to quantify their impact on DOSN sustainability, beyond functional-correctness evaluations in controlled settings.Source: §2 · p. 6

3 Approach

• SuperNova uses best-effort storage to increase data availability through replication across friendship and stranger links, with super-peers helping identify strangers.The goal is increased, not guaranteed 24/7, availability.Source: §3 · p. 6

3.1 Definitions of Different Entities

• A DOSN node has a friend list, and some friends may act as storekeepers by agreeing to hold another user’s data.Source: §3.1 · p. 6

Profile:

• A user profile represents the user and includes data that may be public, restricted to some friends, or private and inaccessible to others.Source: Profile · p. 6

Friendlist:

• A friendlist is the collection of linked user profiles, with friendship assumed to be symmetric.Users add friends based on acquaintances or similarity of interests.Source: Friendlist · p. 7

Storekeepers:

• Storekeepers replicate a user’s data so friends can access it when that user is down; unlike friendship, storekeeping is not necessarily symmetric.New users without enough friends can ask super-peers to store their data.Source: Storekeepers · p. 7

Super-peers:

• Super-peers provide storage to new users who lack enough friends and cooperate to manage services such as maintaining a user list.Different super-peers may specialize in storage and bandwidth, bookkeeping, recommendations, or matchmaking; incentives and reputation mechanisms are envisioned to support accountability and fair rewards.Source: Super-peers · p. 7

Communities:

• Community creators moderate communities and are responsible for storing community data and resolving disputes, with ownership shareable for load balancing.Members may also share bulk community data, while moderators manage all data, including old and new content.Source: Communities · pp. 7–8

Synchronization:

• Logical clocks synchronize nodes with friends, storekeepers, and community members, and help resolve multiple writes and updates.Source: Synchronization · p. 8

Services/Information:

• Public services are available to anyone, while protected services restrict community information to group members.Source: Services/Information · p. 8

List of Directories:

• The system maintains user, super-peer, and community directories, with maintenance shared among super-peers and community owners.The user list is maintained by super-peers, the super-peer list by super-peers themselves, and the community list jointly by community owners and super-peers.Source: List of Directories · p. 8

3.2 Bootstrap of new node

• A new user creates a profile in the Initial Phase, with public profile attributes added to the super-peer-maintained user list.The phase may last from hours to days; username, location, and interests are assumed to be public, though that assumption is not necessary.Source: §3.2 · p. 8
• Super-peers advertise storage capacity, agreement duration, availability times, acceptable content types, and possible advertising terms.Source: §3.2 · pp. 8–9
• During the Take Care Phase, a super-peer stores the new user's replicated data under an agreement, and may delegate storage to nodes that owe it.Stored data is encrypted if necessary so storing parties cannot read or modify it.Source: §3.2 · p. 9
• While receiving super-peer support, users search acquaintances, out-of-channel contacts, and community members for friends who could serve as storekeepers.Source: §3.2 · p. 9
• Super-peers track users' location and uptime, share time-track and stranger-pool information, and can suggest stranger storekeepers when friends do not provide enough availability.Users provide their own uptime and desired coverage intervals when requesting suggestions; they may later relinquish super-peer services once they have enough storekeepers.Source: §3.2 · pp. 9–10
• A new node can remain in the Take Care Phase if it lacks enough independent storekeepers, while each storekeeper stores only a limited amount of the user's latest data.The paper notes that users cannot compel friends or acquaintances to store data.Source: §3.2 · pp. 9–10

3.3 Data Synchronization and Updates.

• The section addresses merging, copying, and updating data between storekeepers.Source: §3.3 · p. 10

Data Storage:

• When an owner reconnects, it requests data from available storekeepers and accepts the latest version; reconnecting storekeepers also synchronize their stored data with other storekeepers or the owner.Source: Data Storage · p. 10
• If storekeepers lack overlapping uptime, a super-peer can temporarily hold the latest data as a rendezvous point until another storekeeper comes online.The design assumes a storekeeper transfers the latest data to another machine before logging off; in this role the super-peer does not handle updates or friends' access requests.Source: Data Storage · p. 10
• While an owner is offline, its friends request the latest data from an available storekeeper, chosen at random among those online.Source: Data Storage · p. 10

Updates:

• If the content owner is online, a friend’s comment triggers an update request to that node.The request includes a logical timestamp.Source: Updates · p. 11
• If the owner is offline, update requests go to all storekeepers that are online.Source: Updates · p. 11
• When multiple updates concern the same item, the owner or storekeepers use timestamps to resolve conflicts.The passage states that every update request comes with a timestamp.Source: Updates · p. 11
• If neither the owner nor any storekeeper is online, comments and posts remain on the commenter’s node until one comes online.Source: Updates · p. 11

3.4 System Model

• The DOSN graph G(N, E) represents users as nodes and mutual friendship links as edges.Source: §3.4 · p. 11
• A node’s storekeepers may include friends and strangers it convinces, while super-peers manage the global stranger pool.A node can share data with a particular super-peer to increase availability.Source: §3.4 · p. 11
• Without copies on other nodes, availability is the time the owner is online; with such copies, availability is defined to include other-node storage.Source: §3.4 · p. 11

4.1 Network

• The network is the 273,798-author giant component of a 2004–2008 DBLP co-authorship graph, with average degree 6 and diameter 23.Source: §4.1 · p. 11

4.2 Trace Generation

• Uptime traces model 12 time zones, six weekday patterns, and four weekend patterns over two weeks, including a scenario with trace changes for 30% of users.In that scenario, 27% receive minor variations and 3% major variations in week two.Source: §4.2 · p. 12

4.3 Various parameters for Storekeepers

• Storekeeper behavior is generalized through Pfd, Psd, and Pas; eight low/high parameter combinations are tested across neighbor, stranger, and super-peer storage strategies.Source: §4.3 · p. 12, §4.4 · pp. 12–13

4.4 Various node strategies for storage

• In the cooperative C2 case, excellent availability is highest and super-peer load lowest; performance is worst in selfish C5.C2 uses Pfd = 50%, Psd = 70%, and Pas = 100%.Source: §4.4 · pp. 13, 15
• Super-peers carry up to 62% load in C5, while acceptance of stranger data (Pas) matters more for their load than stranger-pool participation (Psd).The maximum load is reported for Row C5, Column Snsp.Source: §4.4 · p. 14
• With no uptime deviation, over 50% of nodes reach 100% availability during friends’ online time (FT), versus under 20% over total time (TT); deviation sharply reduces FT performance.Under deviation, all nodes still have availability above 40%; TT performance changes little.Source: §4.4 · p. 15, Figure 2 · p. 16
• In the flat scheme, bootstrap makes it difficult for newcomers to find friends or trusted stranger storekeepers; few nodes exceed 60% availability.The authors compare the flat and super-peer architectures under deviation and non-deviation.Source: Figure 3 · p. 17, §4.4 · p. 17
• Super-peers are not individually online 24/7, so their downtime can reduce availability for many dependent nodes; worst-case availability is substantially below the cooperative best case.In the worst case, availability is 60% or less for 60% of nodes, and around 50% have availability of 70% or less even without deviation.Source: §4.4 · p. 18

• SuperNova gives users flexibility to choose where their data is stored and whose data they store.Source: §5 · p. 18
• The paper evaluates how different user strategy mixes affect DOSN performance and sustainability in a static system.The authors distinguish this quantitative evaluation from prior work focused on enabling technologies.Source: §5 · p. 18
• Future work includes studying dynamic reputation and population changes, adding incentives, refining security, and integrating SuperNova mechanisms into PeerSoN.Source: §5 · p. 18