arXiv subjects

Ruben Mayer

Publications and source records attributed to Ruben Mayer.

48 records · Page 3Linked to original sources

arXiv2018

HYPE: Massive Hypergraph Partitioning with Neighborhood Expansion

Many important real-world applications-such as social networks or distributed data bases-can be modeled as hypergraphs. In such a model, vertices represent entities-such as users or data records-whereas hyperedges model a group membership of the vertices-such as the authorship in a specific topic or the membership of a data record in a specific replicated shard. To optimize such applications, we need an efficient and effective solution to the NP-hard balanced k-way hypergraph partitioning problem. However, existing hypergraph partitioners that scale to very large graphs do not effectively exploit the hypergraph structure when performing the partitioning decisions. We propose HYPE, a hypergraph partitionier that exploits the neighborhood relations between vertices in the hypergraph using an efficient implementation of neighborhood expansion. HYPE improves partitioning quality by up to 95% and reduces runtime by up to 39% compared to streaming partitioning.

cs.DC↗

arXiv2018

P4CEP: Towards In-Network Complex Event Processing

In-network computing using programmable networking hardware is a strong trend in networking that promises to reduce latency and consumption of server resources through offloading to network elements (programmable switches and smart NICs). In particular, the data plane programming language P4 together with powerful P4 networking hardware has spawned projects offloading services into the network, e.g., consensus services or caching services. In this paper, we present a novel case for in-network computing, namely, Complex Event Processing (CEP). CEP processes streams of basic events, e.g., stemming from networked sensors, into meaningful complex events. Traditionally, CEP processing has been performed on servers or overlay networks. However, we argue in this paper that CEP is a good candidate for in-network computing along the communication path avoiding detouring streams to distant servers to minimize communication latency while also exploiting processing capabilities of novel networking hardware. We show that it is feasible to express CEP operations in P4 and also present a tool to compile CEP operations, formulated in our P4CEP rule specification language, to P4 code. Moreover, we identify challenges and problems that we have encountered to show future research directions for implementing full-fledged in-network CEP systems.

cs.NI↗

arXiv2018

ADWISE: Adaptive Window-based Streaming Edge Partitioning for High-Speed Graph Processing

In recent years, the graph partitioning problem gained importance as a mandatory preprocessing step for distributed graph processing on very large graphs. Existing graph partitioning algorithms minimize partitioning latency by assigning individual graph edges to partitions in a streaming manner --- at the cost of reduced partitioning quality. However, we argue that the mere minimization of partitioning latency is not the optimal design choice in terms of minimizing total graph analysis latency, i.e., the sum of partitioning and processing latency. Instead, for complex and long-running graph processing algorithms that run on very large graphs, it is beneficial to invest more time into graph partitioning to reach a higher partitioning quality --- which drastically reduces graph processing latency. In this paper, we propose ADWISE, a novel window-based streaming partitioning algorithm that increases the partitioning quality by always choosing the best edge from a set of edges for assignment to a partition. In doing so, ADWISE controls the partitioning latency by adapting the window size dynamically at run-time. Our evaluations show that ADWISE can reach the sweet spot between graph partitioning latency and graph processing latency, reducing the total latency of partitioning plus processing by up to 23-47 percent compared to the state-of-the-art.

cs.DC↗

arXiv2018

Q-Graph: Preserving Query Locality in Multi-Query Graph Processing

Arising user-centric graph applications such as route planning and personalized social network analysis have initiated a shift of paradigms in modern graph processing systems towards multi-query analysis, i.e., processing multiple graph queries in parallel on a shared graph. These applications generate a dynamic number of localized queries around query hotspots such as popular urban areas. However, existing graph processing systems are not yet tailored towards these properties: The employed methods for graph partitioning and synchronization management disregard query locality and dynamism which leads to high query latency. To this end, we propose the system Q-Graph for multi-query graph analysis that considers query locality on three levels. (i) The query-aware graph partitioning algorithm Q-cut maximizes query locality to reduce communication overhead. (ii) The method for synchronization management, called hybrid barrier synchronization, allows for full exploitation of local queries spanning only a subset of partitions. (iii) Both methods adapt at runtime to changing query workloads in order to maintain and exploit locality. Our experiments show that Q-cut reduces average query latency by up to 57 percent compared to static query-agnostic partitioning algorithms.

cs.DB↗

arXiv2017

The TensorFlow Partitioning and Scheduling Problem: It's the Critical Path!

State-of-the-art data flow systems such as TensorFlow impose iterative calculations on large graphs that need to be partitioned on heterogeneous devices such as CPUs, GPUs, and TPUs. However, partitioning can not be viewed in isolation. Each device has to select the next graph vertex to be executed, i.e., perform local scheduling decisions. Both problems, partitioning and scheduling, are NP-complete by themselves but have to be solved in combination in order to minimize overall execution time of an iteration. In this paper, we propose several heuristic strategies to solve the partitioning and scheduling problem in TensorFlow. We simulate the performance of the proposed strategies in heterogeneous environments with communication-intensive workloads that are common to TensorFlow. Our findings indicate that the best partitioning and scheduling heuristics are those that focus on minimizing the execution time of the critical path in the graph. Those strategies provide a speed-up of up to 4 times in comparison to strategies that are agnostic to the critical path, such as hash-based partitioning and FIFO scheduling.

cs.DC↗

arXiv2017

Knowledge is at the Edge! How to Search in Distributed Machine Learning Models

With the advent of the Internet of Things and Industry 4.0 an enormous amount of data is produced at the edge of the network. Due to a lack of computing power, this data is currently send to the cloud where centralized machine learning models are trained to derive higher level knowledge. With the recent development of specialized machine learning hardware for mobile devices, a new era of distributed learning is about to begin that raises a new research question: How can we search in distributed machine learning models? Machine learning at the edge of the network has many benefits, such as low-latency inference and increased privacy. Such distributed machine learning models can also learn personalized for a human user, a specific context, or application scenario. As training data stays on the devices, control over possibly sensitive data is preserved as it is not shared with a third party. This new form of distributed learning leads to the partitioning of knowledge between many devices which makes access difficult. In this paper we tackle the problem of finding specific knowledge by forwarding a search request (query) to a device that can answer it best. To that end, we use a entropy based quality metric that takes the context of a query and the learning quality of a device into account. We show that our forwarding strategy can achieve over 95% accuracy in a urban mobility scenario where we use data from 30 000 people commuting in the city of Trento, Italy.

cs.DC↗

arXiv2017

FogStore: Toward a Distributed Data Store for Fog Computing

Stateful applications and virtualized network functions (VNFs) can benefit from state externalization to increase their reliability, scalability, and inter-operability. To keep and share the externalized state, distributed data stores (DDSs) are a powerful tool allowing for the management of classical trade-offs in consistency, availability and partitioning tolerance. With the advent of Fog and Edge Computing, stateful applications and VNFs are pushed from the data centers toward the network edge. This poses new challenges on DDSs that are tailored to a deployment in Cloud data centers. In this paper, we propose two novel design goals for DDSs that are tailored to Fog Computing: (1) Fog-aware replica placement, and (2) context-sensitive differential consistency. To realize those design goals on top of existing DDSs, we propose the FogStore system. FogStore manages the needed adaptations in replica placement and consistency management transparently, so that existing DDSs can be plugged into the system. To show the benefits of FogStore, we perform a set of evaluations using the Yahoo Cloud Serving Benchmark.

cs.DC↗

arXiv2017

EmuFog: Extensible and Scalable Emulation of Large-Scale Fog Computing Infrastructures

The diversity of Fog Computing deployment models and the lack of publicly available Fog infrastructure makes the design of an efficient application or resource management policy a challenging task. Such research often requires a test framework that facilitates the experimental evaluation of an application or protocol design in a repeatable and controllable manner. In this paper, we present EmuFog---an extensible emulation framework tailored for Fog Computing scenarios---that enables the from-scratch design of Fog Computing infrastructures and the emulation of real applications and workloads. EmuFog enables researchers to design the network topology according to the use-case, embed Fog Computing nodes in the topology and run Docker-based applications on those nodes connected by an emulated network. Each of the sub-modules of EmuFog are easily extensible, although EmuFog provides a default implementation for each of them. The scalability and efficacy of EmuFog are evaluated both on synthetic and real-world network topologies.

cs.DC↗

arXiv2017

SPECTRE: Supporting Consumption Policies in Window-Based Parallel Complex Event Processing

Distributed Complex Event Processing (DCEP) is a paradigm to infer the occurrence of complex situations in the surrounding world from basic events like sensor readings. In doing so, DCEP operators detect event patterns on their incoming event streams. To yield high operator throughput, data parallelization frameworks divide the incoming event streams of an operator into overlapping windows that are processed in parallel by a number of operator instances. In doing so, the basic assumption is that the different windows can be processed independently from each other. However, consumption policies enforce that events can only be part of one pattern instance; then, they are consumed, i.e., removed from further pattern detection. That implies that the constituent events of a pattern instance detected in one window are excluded from all other windows as well, which breaks the data parallelism between different windows. In this paper, we tackle this problem by means of speculation: Based on the likelihood of an event's consumption in a window, subsequent windows may speculatively suppress that event. We propose the SPECTRE framework for speculative processing of multiple dependent windows in parallel. Our evaluations show an up to linear scalability of SPECTRE with the number of CPU cores.

cs.DC↗

arXiv2017

StreamLearner: Distributed Incremental Machine Learning on Event Streams: Grand Challenge

Today, massive amounts of streaming data from smart devices need to be analyzed automatically to realize the Internet of Things. The Complex Event Processing (CEP) paradigm promises low-latency pattern detection on event streams. However, CEP systems need to be extended with Machine Learning (ML) capabilities such as online training and inference in order to be able to detect fuzzy patterns (e.g., outliers) and to improve pattern recognition accuracy during runtime using incremental model training. In this paper, we propose a distributed CEP system denoted as StreamLearner for ML-enabled complex event detection. The proposed programming model and data-parallel system architecture enable a wide range of real-world applications and allow for dynamically scaling up and out system resources for low-latency, high-throughput event processing. We show that the DEBS Grand Challenge 2017 case study (i.e., anomaly detection in smart factories) integrates seamlessly into the StreamLearner API. Our experiments verify scalability and high event throughput of StreamLearner.

cs.DC↗

arXiv2017

Minimizing Communication Overhead in Window-Based Parallel Complex Event Processing

Distributed Complex Event Processing has emerged as a well-established paradigm to detect situations of interest from basic sensor streams, building an operator graph between sensors and applications. In order to detect event patterns that correspond to situations of interest, each operator correlates events on its incoming streams according to a sliding window mechanism. To increase the throughput of an operator, different windows can be assigned to different operator instances---i.e., identical operator copies---which process them in parallel. This implies that events that are part of multiple overlapping windows are replicated to different operator instances. The communication overhead of replicating the events can be reduced by assigning overlapping windows to the same operator instance. However, this imposes a higher processing load on the single operator instance, possibly overloading it. In this paper, we address the trade-off between processing load and communication overhead when assigning overlapping windows to a single operator instance. Controlling the trade-off is challenging and cannot be solved with traditional reactive methods. To this end, we propose a model-based batch scheduling controller building on prediction. Evaluations show that our approach is able to significantly save bandwidth, while keeping a user-defined latency bound in the operator instances.

cs.DC↗

arXiv2017

The Fog Makes Sense: Enabling Social Sensing Services With Limited Internet Connectivity

Social sensing services use humans as sensor carriers, sensor operators and sensors themselves in order to provide situation-awareness to applications. This promises to provide a multitude of benefits to the users, for example in the management of natural disasters or in community empowerment. However, current social sensing services depend on Internet connectivity since the services are deployed on central Cloud platforms. In many circumstances, Internet connectivity is constrained, for instance when a natural disaster causes Internet outages or when people do not have Internet access due to economical reasons. In this paper, we propose the emerging Fog Computing infrastructure to become a key-enabler of social sensing services in situations of constrained Internet connectivity. To this end, we develop a generic architecture and API of Fog-enabled social sensing services. We exemplify the usage of the proposed social sensing architecture on a number of concrete use cases from two different scenarios.

cs.DC↗