Emrah Cem

Flat and hierarchical epidemics in P2P systems: Energy cost models and analysis

Abstract In large scale distributed systems, epidemic or gossip-based communication mechanisms are preferred for their ease of deployment, simplicity, robustness against failures, load-balancing and limited resource usage. Although they have extensive applicability, there is no prior work on developing energy cost models for epidemic distributed mechanisms. In this study, we address power awareness features of two main groups of epidemics, namely flat and hierarchical. We propose a dominating-set based and power-aware hierarchical epidemic approach that eliminates a significant number of peers from gossiping. To the best of our knowledge, using a dominating set to build a hierarchy for epidemic communication and provide energy efficiency in P2P systems is a novel approach. We develop energy cost model formulations for flat and hierarchical epidemics. In contrast to the prior works, our study is the first one that proposes energy cost models for generic peers using epidemic communication, and examines the effect of protocol parameters to characterize energy consumption. As a case study protocol, we use our epidemic protocol ProFID for frequent items discovery in P2P systems. By means of extensive large scale simulations on PeerSim, we analyze the effect of protocol parameters on energy consumption, compare flat and hierarchical epidemic approaches for efficiency, scalability, and applicability as well as investigate their resilience under realistic churn.

Stepwise fair-share buffering for gossip-based peer-to-peer data dissemination

We consider buffer management in support of large-scale gossip-based peer-to-peer data dissemination protocols. Coupled with an efficient buffering mechanism, system-wide buffer usage can be optimized while providing reliability and scalability in such protocols. We propose a novel approach, stepwise fair-share buffering, that provides uniform load distribution and reduces the overall buffer usage where every peer has a partial view of the system. We report and discuss the comparative performance results with existing buffering approaches as well as random buffering which serves as a benchmark. We present separate evaluations of bufferer selection and gossip-based data dissemination. Reliability, content dissemination time, message delay, buffering delay, and minimum buffer requirements are considered as the key metrics investigated through simulations. The performance of our approach in the case of multiple senders, link failures with multiple bufferers, and scalability to larger networks are investigated. Several power-law and hierarchical overlay topologies are considered. Analytical bounds for reliability of dissemination are also provided.

Impact of sampling design in estimation of graph characteristics

Studying structural and functional characteristics of large scale graphs (or networks) has been a challenging task due to the related computational overhead. Hence, most studies consult to sampling to gather necessary information to estimate various features of these big networks. On the other hand, using a best effort approach to graph sampling within the constraints of an application domain may not always produce accurate estimates. In fact, the mismatch between the characteristics of interest and the utilized network sampling methodology may result in incorrect inferences about the studied characteristics of the underlying system. In this study we empirically investigate the sources of information loss in a sampling process; identify the fundamental factors that need to be carefully considered in a sampling design; and use several synthetic and real world graphs to elaborately demonstrate the mismatch between the sampling design and graph characteristics of interest.

ProFID: Practical frequent items discovery in peer-to-peer networks

We address the problem of discovering frequent items in unstructured P2P networks which is relevant for several distributed services such as cache management, data replication, query refinement, topology optimization and security. This study makes the following contributions to the current state of the art. First, we propose and develop a fully distributed Protocol for Frequent Items Discovery (ProFID) where the result is produced at every peer. ProFID uses gossip-based (epidemic) communication, a novel pairwise averaging function and system size estimation together to discover frequent items in an unstructured P2P network. We also propose a practical rule for convergence of the algorithm. In contrast to the previous works, each peer gives a local decision for convergence based on the change of updated local state. We developed a model of ProFID in PeerSim and performed various experiments to compare and evaluate its efficiency, scalability, and applicability. The protocols resilience under realistic churn models was studied. For evaluating the effect of network dynamics, we deployed our protocol on the Internet-scale real network PlanetLab. We also compared the accuracy and scalability of ProFID with the adaptive Push-Sum algorithm. Our results confirm the practical nature, ease of deployment and efficiency of our approach, and also show that it outperforms adaptive Push-Sum in terms of accuracy, convergence speed and message overhead.

Energy Cost Model for Frequent Item Set Discovery in Unstructured P2P Networks

For large scale distributed systems, designing energy efficient protocols and services has become as significant as considering conventional performance criteria like scalability, reliability, fault-tolerance and security. We consider frequent item set discovery problem in this context. Although it has attracted attention due to its extensive applicability in diverse areas, there is no prior work on energy cost model for such distributed protocols. In this paper, we develop an energy cost model for frequent item set discovery in unstructured P2P networks. To the best of our knowledge, this is the first study that proposes an energy cost model for a generic peer using gossip-based communication. As a case study protocol, we use our gossip-based approach ProFID for frequent item set discovery. After developing the energy cost model, we examine the effect of protocol parameters on energy consumption using our simulation model on PeerSim and compare push–pull method of ProFID with the well-known push-based gossiping approach. Based on the analysis results, we reformulate the upper bound for the peer’s energy cost.

ProFID: Practical Frequent Item Set Discovery in Peer-to-Peer Networks

This study addresses the problem of discovering frequent items in unstructured P2P networks. We propose a fully distributed Protocol for Frequent Item set Discovery (ProFID) where the result is produced at every peer. We also propose a practical rule for convergence of the algorithm. Finally, we evaluate the efficiency of our approach through an extensive simulation study on PeerSim.

Average degree estimation under ego-centric sampling design

Estimating the structural characteristics of large graphs from a sample is a classical problem. In this study, we propose asymptotically unbiased estimators for the average degree characteristic of a network under ego-centric sampling. In this sampling design, we first sample a number of vertices called ego vertices from the underlying graph and then obtain their ego-centric graph. Ego-centric graph of a sampled vertex is defined as the subgraph induced by the vertices within 1-hop neighborhood of the sampled ego vertex. We compare the proposed estimators with the estimator that do not utilize the neighborhood information using both real-world and synthetic large-scale graphs. The results show that utilization of the neighborhood information does not always increase the estimation accuracy depending on the sampling budget usage and the structure of the underlying graph.

Estimation of structural properties of online social networks at the extreme

Sampling is a commonly used technique for studying structural properties of online social networks (OSNs). Due to privacy, business, and performance concerns, OSN service providers impose limitations on data access for third parties. The implication of this practice is that one needs to come up with an applicable sampling scheme that can function under these limitations to efficiently estimate structural properties of interest. In this paper, we study how accurately some important properties of graphs can be estimated under a limited data access model. More specifically, we consider random neighbor access (RNA) model as a rather limited data access model in OSNs. In the RNA model, the only query available to get data from the studied graph is the random neighbor query which returns the id of a random neighbor for a given vertex id. We propose various sampling schemes and estimators for average degree and network size under the RNA model. We conduct extensive experiments on both real world OSN graphs and synthetic graphs (1) to measure the performance of the proposed estimators and (2) to identify the factors affecting the accuracy of our estimators. We find that while the average degree estimators can make accurate estimations with reasonable sample sizes despite the extreme data access limitations of the RNA model, network size estimators require quite large sample sizes for accurate estimations.

Estimating the size and average degree of online social networks at the extreme

Given the increasingly limiting nature of online social networks (OSNs), studying their structural characteristics under a limited data access model becomes important. In this study, we propose estimators for network size and average degree characteristics of OSNs. We sample an OSN graph using random neighbor API calls. A random neighbor API call returns only the id of a randomly selected neighbor of a given user. Although the existing estimators give good accuracy estimations for a given sample size, they are not applicable under the extremely limited data access model considered here. We conduct experiments on real world graphs to measure the performance of the proposed estimators.

A distributed approach for computing sum aggregation in P2P networks

In large-scale peer-to-peer (P2P) systems, epidemic or gossip-based protocols have become significant as an alternative to the hierarchical protocols due to their simplicity, robustness and scalability. In this study, we provide the details of the sum aggregation technique, namely atomic pairwise averaging, used in peer-to-peer networks. Various distributed services can utilize sum aggregation as a functional building. In contrast to previous studies, sum aggregate is computed indirectly by computing average aggregate along with the network size. By this technique, the requirement of checking if two peers have exchanged states before is eliminated. Moreover, analytical discussion of why the atomic pairwise averaging technique results in the average aggregate value of items at each peer is provided. Furthermore, simulation results of effects of network, as well as the effects of parameters are presented.