Date of Award
Doctor of Philosophy (PhD)
Dr. Haiying Shen, Committee Chair
Dr. Richard R. Brooks
Dr. Kuang-Ching (KC) Wang
Dr. Feng Luo
In the past few years, Online Social Networks (OSNs) have dramatically spread over the world. Facebook , one of the largest worldwide OSNs, has 1.35 billion users, 82.2% of whom are outside the US . The browsing and posting interactions (text content) between OSN users lead to user data reads (visits) and writes (updates) in OSN datacenters, and Facebook now serves a billion reads and tens of millions of writes per second . Besides that, Facebook has become one of the top Internet traﬃc sources  by sharing tremendous number of large multimedia ﬁles including photos and videos. The servers in datacenters have limited resources (e.g. bandwidth) to supply latency eﬃcient service for multimedia ﬁle sharing among the rapid growing users worldwide. Most online applications operate under soft real-time constraints (e.g., ≤ 300 ms latency) for good user experience, and its service latency is negatively proportional to its income. Thus, the service latency is a very important requirement for Quality of Service (QoS) to the OSN as a web service, since it is relevant to the OSN’s revenue and user experience. Also, to increase OSN revenue, OSN service providers need to constrain capital investment, operation costs, and the resource (bandwidth) usage costs. Therefore, it is critical for the OSN to supply a guaranteed QoS for both text and multimedia contents to users while minimizing its costs. To achieve this goal, in this dissertation, we address three problems. i) Data distribution among datacenters: how to allocate data (text contents) among data servers with low service latency and minimized inter-datacenter network load; ii) Eﬃcient multimedia ﬁle sharing: how to facilitate the servers in datacenters to eﬃciently share multimedia ﬁles among users; iii) Cost minimized data allocation among cloud storages: how to save the infrastructure (datacenters) capital investment and operation costs by leveraging commercial cloud storage services. Data distribution among datacenters. To serve the text content, the new OSN model, which deploys datacenters globally, helps reduce service latency to worldwide distributed users and release the load of the existing datacenters. However, it causes higher inter-datacenter communica-tion load. In the OSN, each datacenter has a full copy of all data, and the master datacenter updates all other datacenters, generating tremendous load in this new model. The distributed data storage, which only stores a user’s data to his/her geographically closest datacenters, simply mitigates the problem. However, frequent interactions between distant users lead to frequent inter-datacenter com-munication and hence long service latencies. Therefore, the OSNs need a data allocation algorithm among datacenters with minimized network load and low service latency. Eﬃcient multimedia ﬁle sharing. To serve multimedia ﬁle sharing with rapid growing user population, the ﬁle distribution method should be scalable and cost eﬃcient, e.g. minimiza-tion of bandwidth usage of the centralized servers. The P2P networks have been widely used for ﬁle sharing among a large amount of users [58, 131], and meet both scalable and cost eﬃcient re-quirements. However, without fully utilizing the altruism and trust among friends in the OSNs, current P2P assisted ﬁle sharing systems depend on strangers or anonymous users to distribute ﬁles that degrades their performance due to user selﬁsh and malicious behaviors. Therefore, the OSNs need a cost eﬃcient and trustworthy P2P-assisted ﬁle sharing system to serve multimedia content distribution. Cost minimized data allocation among cloud storages. The new trend of OSNs needs to build worldwide datacenters, which introduce a large amount of capital investment and maintenance costs. In order to save the capital expenditures to build and maintain the hardware infrastructures, the OSNs can leverage the storage services from multiple Cloud Service Providers (CSPs) with existing worldwide distributed datacenters [30, 125, 126]. These datacenters provide diﬀerent Get/Put latencies and unit prices for resource utilization and reservation. Thus, when se-lecting diﬀerent CSPs’ datacenters, an OSN as a cloud customer of a globally distributed application faces two challenges: i) how to allocate data to worldwide datacenters to satisfy application SLA (service level agreement) requirements including both data retrieval latency and availability, and ii) how to allocate data and reserve resources in datacenters belonging to diﬀerent CSPs to minimize the payment cost. Therefore, the OSNs need a data allocation system distributing data among CSPs’ datacenters with cost minimization and SLA guarantee. In all, the OSN needs an eﬃcient holistic data distribution and storage solution to minimize its network load and cost to supply a guaranteed QoS for both text and multimedia contents. In this dissertation, we propose methods to solve each of the aforementioned challenges in OSNs. Firstly, we verify the beneﬁts of the new trend of OSNs and present OSN typical properties that lay the basis of our design. We then propose Selective Data replication mechanism in Distributed Datacenters (SD3) to allocate user data among geographical distributed datacenters. In SD3,a datacenter jointly considers update rate and visit rate to select user data for replication, and further atomizes a user’s diﬀerent types of data (e.g., status update, friend post) for replication, making sure that a replica always reduces inter-datacenter communication. Secondly, we analyze a BitTorrent ﬁle sharing trace, which proves the necessity of proximity-and interest-aware clustering. Based on the trace study and OSN properties, to address the second problem, we propose a SoCial Network integrated P2P ﬁle sharing system for enhanced Eﬃciency and Trustworthiness (SOCNET) to fully and cooperatively leverage the common-interest, geographically-close and trust properties of OSN friends. SOCNET uses a hierarchical distributed hash table (DHT) to cluster common-interest nodes, and then further clusters geographically close nodes into a subcluster, and connects the nodes in a subcluster with social links. Thus, when queries travel along trustable social links, they also gain higher probability of being successfully resolved by proximity-close nodes, simultaneously enhancing eﬃciency and trustworthiness. Thirdly, to handle the third problem, we model the cost minimization problem under the SLA constraints using integer programming. According to the system model, we propose an Eco-nomical and SLA-guaranteed cloud Storage Service (ES3), which ﬁnds a data allocation and resource reservation schedule with cost minimization and SLA guarantee. ES3 incorporates (1) a data al-location and reservation algorithm, which allocates each data item to a datacenter and determines the reservation amount on datacenters by leveraging all the pricing policies; (2) a genetic algorithm based data allocation adjustment approach, which makes data Get/Put rates stable in each data-center to maximize the reservation beneﬁt; and (3) a dynamic request redirection algorithm, which dynamically redirects a data request from an over-utilized datacenter to an under-utilized datacenter with suﬃcient reserved resource when the request rate varies greatly to further reduce the payment. Finally, we conducted trace driven experiments on a distributed testbed, PlanetLab, and real commercial cloud storage (Amazon S3, Windows Azure Storage and Google Cloud Storage) to demonstrate the eﬃciency and eﬀectiveness of our proposed systems in comparison with other systems. The results show that our systems outperform others in the network savings and data distribution eﬃciency.
Liu, Guoxin, "An Efficient Holistic Data Distribution and Storage Solution for Online Social Networks" (2015). All Dissertations. 1774.