Glauber D. Gonçalves

Modeling the Dropbox client behavior

Cloud storage systems are currently very popular, generating a large amount of traffic. Indeed, many companies offer this kind of service, including worldwide providers such as Dropbox, Microsoft and Google. These companies, as well as new providers entering the market, could greatly benefit from knowing typical workload patterns that their services have to face in order to develop more cost-effective solutions. However, despite recent analyses of typical usage patterns and possible performance bottlenecks, no previous work investigated the underlying client processes that generate workload to the system. In this context, this paper proposes a hierarchical two-layer model for representing the Dropbox client behavior. We characterize the statistical parameters of the model using passive measurements gathered in 3 different network vantage points. Our contributions can be applied to support the design of realistic synthetic workloads, thus helping in the development and evaluation of new, well-performing personal cloud storage services.

Characterizing scholar popularity: a case study in the computer science research community

A common live debate among scholars regards the popularity, productivity and impact of research. This paper aims to contribute to such discussion by quantifying the impact of various academic features on a scholar popularity throughout her career. Using a list of over 2 million publications in the Computer Science research area obtained from two large digital libraries, we analyze how features that capture the number and rate of publications, number and quality of publication venues, and the importance of the scholar in the co-authorship network relate to the scholar popularity. We also investigate the temporal dynamics of scholar popularity, identifying a few common profiles, and characterizing scholars in each profile according to their academic features.

SopCast P2P live streaming: live session traces and analysis

P2P-TV applications have attracted a lot of attention from the research community in the last years. Such systems generate a large amount of data which impacts the network performance. As a natural consequence, characterizing these systems has become a very important task to develop better multimedia systems. However, crawling data from P2P live streaming systems is particularly challenging by the fact that most of these applications have private protocols. In this work, we present a set of logs from a very popular P2P live streaming application, the SopCast. We describe our crawling methodology, and present a brief SopCast characterization. We believe that our logs and the characterization can be used as a starting point to the development of new live streaming systems.

Using centrality metrics to predict peer cooperation in live streaming applications

The lack of cooperation in Peer-to-Peer (P2P) applications poses serious challenges to the quality of service provided to their clients, specifically in P2P live streaming applications given their strict real-time constraints. We here investigate the potential of exploiting topological properties of the P2P overlay network to predict the level of cooperation of a peer, measured by the ratio of the upload to the download traffic during a pre-defined time window. Using data collected from SopCast, we first show that centrality metrics provide good evidence of a peers cooperation level in the system. We then develop a regression-based model that is able to estimate, with reasonable accuracy, the level of cooperation of a peer in the near future given its centrality measures in the recent past. Our proposed strategy complements existing incentive mechanisms for cooperation in P2P live streaming, and can be applied to detect non-cooperative peers.

Characterizing interactions among members of deaf communities in orkut

This paper presents a characterization of interaction phenomena among members of communities aimed at deaf people in an online social network system, namely Orkut. The results revealed that members of deaf communities are tighter than members of other communities. However, analysis of the interface indicates that it does not always address their specific needs.

Analyzing the Impact of Dropbox Content Sharing on an Academic Network

Cloud storage services (e.g., Dropbox) are a popular means for sharing content and performing collaborative work. Yet, content sharing via cloud might result in bandwidth wastage when repetitive data is downloaded by different users in the same network domain. This paper first characterizes sharing patterns in Dropbox by analyzing data collected from a campus network for 4 months. We identify that the volume of data sharing in such homogeneous environment is reasonably high. Next, we use the characterization results to implement a synthetic workload generator that allows us to test alternatives for the Dropbox synchronization protocol. We then propose a synchronization architecture that includes network caches to temporarily hold user updates. Our evaluation of the proposed solution indicates that, even with a small cache, it is possible to achieve almost the maximum possible reduction of downloads from remote servers, thus benefiting storage providers, end users and the Internet.

Characterizing and Modeling the Dropbox Workload

This paper presents a characterization and modeling of the workload of the currently most popular cloud storage system, Drop box. The workload is analyzed from two complementary perspectives. On one side, characteristics of the Drop box folders of a set of 333 volunteers, including number, type and sizes of files, replication level, and file update dynamics are analyzed. On the other side, the client behavior patterns are modeled based on the analysis of Drop box traffic collected from 3 universities. Our results can be explored to design realistic synthetic workloads, which in turn can support the design and evaluation of more cost-effective cloud storage services.

The Impact of Content Sharing on Cloud Storage Bandwidth Consumption

Cloud storage offers a popular environment for users to store and share content. Yet, content sharing leads to multiple downloads of the same content when users synchronize devices. These downloads contribute to bandwidth waste and increase server workloads. To what extent does this occur? Would network caches alleviate the problem? The authors address these questions by investigating the traffic generated by Dropbox, a popular cloud storage service. Using data collected from four networks, they show that a large fraction (57-70 percent) of downloads generated by Dropbox users is associated with content shared among multiple devices. Moreover, despite optimizations such as the LAN Sync protocol, up to 25 percent of Dropbox download traffic comes from content replicas. The authors then evaluate an alternative synchronization architecture that uses caches to offload storage servers from such downloads. Their experiments show that the approach cost-effectively avoids most repetitive downloads, benefiting service providers, the network, and end users.