Program

The track provides a forum for scientists and engineers in academia and industry to present and discuss their latest research findings on selected topics in dependable and adaptive distributed systems.

Wed, March 20, 14:10-15:50

• Improving Context Interpretation by Using Fuzzy Policies: The Case of Adaptive Video Streaming
  Lucas Provensi, Frank Eliassen, Roman Vitenberg and Romain Rouvoy

  Adaptation is an increasingly important requirement for software systems executing in large-scale, heterogeneous, and dynamic environments. A central aspect of the adaptation methodology is management of contextual information needed to support the adaptation process. A major design challenge of managing contextual data lies in the fact that the information is partial, uncertain, and inherently suitable for diverging interpretations. While existing adaptation solutions focus on techniques, methods, and tools, the challenge of managing and interpreting ambiguous contextual information remains largely unresolved. In this paper we present a new adaptation approach that aims to overcome these issues by applying fuzzy set theory and approximate reasoning. We have defined a knowledge management scheme that allows the interpretation of imprecise information and effectively integrated it into the adaptation feedback control loop. To test and evaluate our solution, we implemented it in an adaptation engine to perform rate control for media streaming applications. We show the benefits of our approach in terms of flexibility and performance when compared to more traditional methods, such as TCP-friendly rate control.

• Hyphen: A Flexible Middleware for Overlay Network Construction
  Mouna Allani, Benoît Garbinato and Peter Pietzuch

  Overlay networks form the core part of peer-to-peer (P2P) applications such as application-level multicast, content distribution and media streaming. To ease development, middleware solutions and toolkit libraries have been proposed in the past to help with the implementation of overlay networks. Existing solutions, however, are either too generic by only providing low-level communication abstractions, requiring developers to implement algorithms for overlay networks from scratch, or too restrictive by only supporting a particular overlay topology with fixed properties. In this paper, we argue that it is possible to find a middle ground between these two extremes. We describe Hyphen, a middleware for overlay construction and maintenance that supports a range of overlay topologies with custom properties, and show how it can replace topology construction for a variety of application-level multicast systems. Unlike previous efforts, Hyphen can construct and maintain a range of overlay topologies such as trees and forests with specific optimisation goals such as low latency or high bandwidth. By using a gossip-based mechanism to define topologies implicitly, Hyphen can scale to many peers and achieve low construction overhead. Our experimental evaluation with Bullet and Splitstream, two P2P streaming systems, shows that Hyphen can construct a bandwidth-optimised tree for Bullet that achieves a higher streaming rate than the original Bullet implementation, and that it can construct a more reliable forest for Splitstream by taking individual peer reliability into account.

• Stheno, a Real-Time Fault-Tolerant P2P Middleware Platform for Light-Train Systems
  Rolando Martins, Luís Lopes, Fernando Silva and Priya Narasimhan

  Large scale information systems, such as public information systems for light-train/metro networks, must be able to fulfill contractualized Service Level Agreements (SLAs) in terms of end-to-end latencies and jitter, even in the presence of faults. Failure to do so has potential legal and financial implications for the software developers. Current middleware solutions have a hard time coping with these demands due, fundamentally, to a lack of adequate, simultaneous, support for fault-tolerance (FT) and real-time (RT) tasks. In this paper we present Stheno, a general purpose peer-to-peer (P2P) middleware system that builds on previous work from TAO and MEAD to provide: (a) configurable, transparent, FT support by taking advantage of the P2P layer topology awareness to efficiently implement Common Of The Shelf (COTS) replication algorithms and replica management strategies, and; (b) kernel-level resource reservation integrated with well-known threading strategies based on priorities to provide more robust support for soft real-time tasks. An evaluation of the first (unoptimized) prototype for the middleware shows that Stheno is able to match and often greatly exceed the SLA agreements provided by our target system, the light-train/metro information system developed and maintained by EFACEC, and currently deployed at multiple cities in Europe and Brazil.

• Understanding the Quality of Experience in Modern Distributed Interactive Multimedia Applications in Presence of Failures: Metrics and Analysis
  Narasimha Raghavan Veeraragavan, Leonardo Montecchi, Nicola Nostro, Andrea Bondavalli, Roman Vitenberg and Hein Meling

  Recent technological advances has made it possible to design bandwidth demanding distributed interactive multimedia applications such as the World Opera. In this application artists physically located at different opera houses across the globe, can participate in a single united performance and interact almost as if they were co-located. In this paper we propose a novel approach for evaluating the performability of distributed multimedia applications in presence of failures. We first define a new metric, perceived reliability, which attempts to capture the user-perceived quality of experience resulting from the possible quality degradation of the World Opera performance. Next, we develop a Stochastic Activity Network model, in which the new metric is evaluated to understand how perceived quality is affected by different stage configurations. The latter provides useful guidelines for system engineers to improve the perceived quality of World Opera performances despite presence of failures.

• Maximizing Availability of Content in Disruptive Environments by Cross-Layer Optimization
  Minyoung Kim, Je-Min Kim, Mark-Oliver Stehr, Ashish Gehani, Dawood Tariq and Jin-Soo Kim

  Emerging applications such as search-and-rescue operations, CNS (communication, navigation, surveillance), smart spaces, vehicular networks, mission-critical infrastructure, and disaster control require reliable content distribution under harsh network conditions and all kinds of component failures. In such scenarios, potentially heterogeneous networked components - where the networks lack reliable connections - need to be managed to improve scalability, performance, and availability of the overall system. Inspired by delay- and disruption-tolerant networking, this paper presents a distributed cross-layer monitoring and optimization method for secure content delivery as a first step toward decentralized content-based mobile ad hoc networking. In particular, we address the availability maximization problem by embedding monitoring and optimization within an existing content-distribution framework. The implications of policies at security, caching, and hardware layers that control in-network storage and hop-by-hop dissemination of content then are analyzed to maximize the content availability in disruptive environments.

Wed, March 20, 16:20-18:00

• MoSQL: An Elastic Storage Engine for MySQL
  Alexander Tomic, Daniele Sciascia and Fernando Pedone

  We present MoSQL, a MySQL storage engine using a transactional distributed key-value store system for atomicity, isolation and durability and a B+Tree for indexing purposes. Despite its popularity, MySQL is still without a general- purpose storage engine providing high availability, serializability, and elasticity. In addition to detailing MoSQL’s design and implementation, we assess its performance with a number of benchmarks which show that MoSQL scales to a fairly large number of nodes on-the-fly, that is, additional nodes can be added to a running instance of the system.

• A Multi-Resource Dynamic Load Balancing Algorithm for Cache Systems
  Yu Jia, Ricardo Jimenez-Peris, Marta Patiño-Martinez, Ivan Brondino and Dianfu Ma

  With the advent of cloud computing model, distributed caches have become the cornerstone for building scalable applications. Popular systems like Facebook or Google App Engine use Memcached, a highly scalable distributed object cache, to speed up applications by avoiding database accesses. Distributed object caches assign objects to cache instances based on a hashing function, and objects are not moved from a cache instance to another unless more instances are added to the cache and objects are redistributed. This may lead to situations where some cache instances are overloaded when some of the objects they store are frequently accessed, while other cache instances are less frequently used. In this paper we propose a multi-resource load balancing algorithm for distributed cache systems. The algorithm aims at balancing both CPU and Memory resources among cache instances by redistributing stored data. Considering the possible conflict of balancing multiple resources at the same time, we give CPU and Memory resources weighted priorities based on the runtime load distributions. A scarcer resource is given a higher weight than a less scarcer resource when load balancing. Based on the monitoring of a node’s utility load, a data unit’s resource consumptions and the system imbalance degree. Besides, since continuous rebalance of the system may affect the QoS of applications utilizing the cache system, our data selection policy ensures that each data migration minimizes the system imbalance degree and hence, the total reconfiguration cost can be minimized. An extensive simulation is conducted to compare our policy with other policies. Our policy shows a significant improvement in time efficiency and decrease in reconfiguration cost.

• Adaptive Monitoring of Web-based Applications: A Performance Study
  João Paulo Magalhães and Luís Moura Silva

  Mean-time-to-repair (MTTR) is an important factor in the assessment of the availability of a system. With applications crossing the enterprise boundaries and relying on numerous layers of technology the ability to detect and pin- point anomalies becomes more complex, leading to a higher MTTR. Combine application-level profiling and runtime analysis may improve these aspects, although some efforts are still necessary to reduce their performance impact. In this paper we study the performance impact induced by different application-level monitoring tools, targeted for the detection and localization of performance anomalies in Web-based applications. Adaptive and selective algorithms, able to self-adapt the monitoring behavior, are proposed to minimize the performance impact induced by application- level profiling. From the experimental results, becomes clear their usefulness to minimize the performance impact without compromising the ability to detect and pinpoint the anomalies. These outcomes are very favorable to the combination of application-level profiling and runtime analysis, as a way to reduce the MTTR in production systems.

• Experience with a Middleware Infrastructure for Service Oriented Financial Applications
  Jose Pedro Oliveira and Jose Pereira

  Financial institutions, acting as financial intermediaries, need to handle numerous information sources and feed them to multiple processing, storage, and display services. This requires filtering and routing, but these feeds are usually provided in custom formats and protocols that are not the best fit for further processing. Moreover, the sheer volume of information and stringent timeliness and reliability requirements make this a substantial task. In this paper, i) we characterize one of these information feeds (the Exchange Data Publisher feed from the NYSE Euronext European Cash Markets) and ii) we present and evaluate a dissemination system for this particular feeder based on commodity hardware and open-source message-oriented middleware (Apache Qpid). This allows us to assess the feasibility of this approach and to point out the main challenges to be overcome.

• Identifying Incompatible Service Implementations using Pooled Decision Trees
  Christian Inzinger, Waldemar Hummer, Benjamin Satzger, Philipp Leitner and Schahram Dustdar

  In this paper we study fault localization techniques for identification of incompatible configurations and implementations in service-based applications (SBAs). Practice has shown that standardized interfaces alone do not guarantee compatibility of services originating from different partners. Hence, dynamic runtime instantiations of such SBAs pose a great challenge to reliability and dependability. The aim of this work is to monitor and analyze successful and faulty executions in SBAs, in order to proactively detect incompatible configurations at runtime. We propose an approach using pooled decision trees for localization of faulty service parameter and binding configurations, explicitly addressing transient and changing fault conditions. The presented fault localization technique works on a per-request basis and is able to take individual service inputs into account. Considering not only the service configuration but also the service input data as a parameters for the fault localization algorithm increases the computational complexity by an order of magnitude. Hence, our performance evaluation is targeted at large-scale SBAs and illustrates the feasibility and decent scalability of the approach.

Poster session, Wed, March 20

• Improving transaction abort rates without compromising throughput through judicious scheduling
  Ana Nunes and José Pereira

  We propose a novel optimistic concurrency control mechanism that reduces transaction abort rate by minimizing the time during which transactions are vulnerable to abort, without compromising throughput. Our contribution is two-fold: we throttle transaction execution with an adaptive mechanism based on the state of the transaction queues; also, we introduce an adaptive transaction scheduling mechanism allowing out-of-order execution based on conflict classes.

• Towards a Ranking Framework for Software Components
  Dhyanesh Chaudhari, Mohammad Zulkernine and Komminist Weldemariam

  Software components are meant to be reusable and flexi- ble by design. These characteristics and others continue attracting software developers to adapt a component (typi- cally designed elsewhere) into their systems. However, soft- ware components are also believed to be most vulnerable to reliability and security problems due to the existence of non- obvious faults. In this paper we present a framework that can help system developers in detecting and ranking compo- nents failures systematically so that more reliable software can be achieved. The approach can allow monitoring critical components within a system under instrumentation, detect- ing failures based on specifications and using failure data and input from developers, and ranking the components ac- cordingly. A prototype tool is designed along with a number of failure scenarios to detect specific failure types within a component. We conducted an experimental evaluation to assess the effectiveness of the proposed framework and to measure its performance overhead. Our experimental re- sults show that the approach can benefit system developers by prioritizing the maintenance of components with a mini- mal overhead.

Track Program Co-Chairs

Important Dates