Dependable, Adaptive, and Secure Distributed Systems17th DADS Track of the
37th ACM Symposium on Applied Computing
16th DADS 2021
15th DADS 2020
14th DADS 2019
13th DADS 2018
12th DADS 2017
11th DADS 2016
10th DADS 2015
9th DADS 2014
8th DADS 2013
7th DADS 2012
6th DADS 2011
5th DADS 2010
4th DADS 2009
3rd DADS 2008
2nd DADS 2007
1st DADS 2006
April 25 - 29, 2022
Brno, Czech Republic
The Symposium on Applied Computing has been a primary gathering forum for applied computer scientists, computer engineers, software engineers, and application developers from around the world. SAC 2022 is sponsored by the ACM Special Interest Group on Applied Computing and the SRC Program is sponsored by Microsoft Research.
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, adaptive and trustworthy distributed systems and services.
Details see SAC program page.
A General and Configurable Framework for Blockchain-based Marketplaces
Andrea Merlina, Roman Vitenberg and Vinay Setty
The first generation of blockchain focused on digital currencies and secure storage, management and transfer of tokenized values. Thereafter, the focus has been shifting from currencies to a broader application space. In this paper, we systematically explore marketplace types and properties, and consider the mechanisms required to support those properties through blockchain. We propose a generic and configurable framework for blockchain-based marketplaces, and describe how popular marketplace types, price discovery policies, and other configuration parameters are implemented within the framework by presenting concrete event-based algorithms. Finally, we consider two use cases with widely diverging properties and show how the proposed framework supports them.
Redundant Dataflow Applications on Clustered Manycore Architectures
Christoph Kühbacher, Theo Ungerer and Sebastian Altmeyer
Increasing performance requirements in the embedded systems domain have encouraged a drift from singlecore to multicore processors. Cars are an example for complex embedded systems in which the use of multicores continues to grow. The requirements of software components running in modern cars are diverse. On the one hand there are safety-critical tasks like the airbag control, on the other hand tasks which do not have any safety-related requirements at all, for example those controlling the infotainment system. Trends like autonomous driving lead to tasks which are simultaneously safety-critical and computationally complex. To satisfy the requirements of modern embedded applications we developed a dataflow-based runtime environment (RTE) for clustered manycore architectures. The RTE is able to execute dataflow graphs in various redundancy configurations and with different schedulers. We implemented our RTE design on the Kalray Bostan Massively Parallel Processor Array and evaluated all possible configurations for three common computation tasks. To classify the performance of our RTE, we compared the non-redundant graph executions with OpenCL versions of the three applications. The results show that our RTE can come close or even surpass Kalray’s OpenCL framework, although maximum performance was not the primary goal of our design.
RANC: Reward-All Nakamoto Consensus
Rami Khalil and Naranker Dulay
In this work we present Reward-All Nakamoto-Consensus (RANC), a Proof-of-Work cryptocurrency that resiliently rewards each miner with a number of coins that is directly proportional to its individual mining power, rather than to its relative share of the entire network’s mining power as done in Bitcoin. Under this approach, the security of mining in RANC achieves near-perfect incentive compatibility, and near-zero censorship susceptibility, for adversarial mining shares up to 45%, but at the cost of regression in subversiongain resilience. Moreover, mining rewards in RANC exhibit significantly lower variance for non-majority miners compared to NC, enabling dependable reward stability. Consequently, depending on the network transaction-fees, RANC improves miner’s waiting time for rewards, and incentivizes forming mining pools smaller than required in Bitcoin for equal reward stability. A detailed specification of RANC is presented, along with an evaluation of the practicality and efficiency achieved by our prototype RANC implementation.
ElastiQuant: Elastic Quantization Strategy for Communication Efficient Distributed Machine Learning in IoT
Bharath Sudharsan, John G. Breslin, Muhammad Intizar Ali, Peter Corcoran and Rajiv Ranjan
In training distributed machine learning, communicating model updates among workers has always been a bottleneck. The magnitude of impact on the quality of resultant models is higher when distributed training on low hardware specification devices and in uncertain real-world IoT networks where congestion, latency, bandwidth issues are common. In this scenario, gradient quantization plus encoding is an effective way to reduce cost when communicating model updates. Other approaches can be to limit the clientserver communication frequency, adaptive compression by varying the spacing between quantization levels, reusing outdated gradients, deep compression to reduce transmission packet size, and adaptive tuning of the number of bits transmitted per round. The optimization levels provided by such and other non-comprehensive approaches do not suffice for high-dimensional NN models with large size model updates. This paper presents ElastiQuant, an elastic quantization strategy that aims to reduce the impact caused by limitations in distributed IoT training scenarios. The distinguishable highlights of this comprehensive work are: (i) theoretical assurances and bounds on variance and number of communication bits are provided, (ii) worst-case variance analysis is performed, and (iii) momentum is considered in convergence assurance. ElastiQuant experimental evaluation and comparison with top schemes by distributed training 5 ResNets on 18 edge GPUs over ImageNet and CIFAR datasets show: improved solution quality in terms of ˜ 2-11 % training loss reduction, ˜ 1-4 % accuracy boost, and ˜ 4-22 % variance drop; positive scalability due to higher communication compression resulting in saving bandwidth and ˜ 4-30 min per epoch training speedups.
An event-driven strategy for reactive replica balancing on Apache Hadoop Distributed File System
Rhauani Fazul and Patrícia Pitthan Barcelos
Distributed file systems are essential to support applications that handle large volumes of data. One of the most widely used distributed file systems is the HDFS, the Apache Hadoop’s Distributed File System. Data replication, which is at the heart of the HDFS storage model, is essential for fault tolerance and performance. As new data is loaded into the system, it is common for the distribution of the replicas among the nodes to become unbalanced. HDFS Balancer is the official solution for data balancing through replica rearrangement. Currently, it is up to the system administrator to monitor the HDFS status and, when considered necessary, run the balancer, which creates a dependency that is inadequate and inefficient in many situations. This work presents an event-based strategy to make the replica balancing process in HDFS transparent. To this end, we have created a metrics observation model and a structure that, based on standardized trigger events, automatically determines when corrective actions should be taken and triggers the reactive balancing process in the file system. The evaluation results demonstrate that the proposed solution is able to keep the cluster balanced, which contributes to overall data reliability and availability. In addition, it allows taking better advantage of data locality during reading operations over the stored data in the HDFS.
Adaptive Database Synchronization for an Online Analytical Cloud-to-Edge Continuum
Daniel Costa, Jose Pereira, Ricardo Vilaça and Nuno Faria
Wide availability of edge computing platforms, as expected in emerging 5G networks, enables a computing continuum between centralized cloud services and the edge of the network, close to end-user devices. This is particularly appealing for online analytics as data collected by devices is made available for decisionmaking. However, cloud-based parallel-distributed data processing platforms are not able to directly access data on the edge. This can be circumvented, at the expense of freshness, with data synchronization that periodically uploads data to the cloud for processing. In this work, we propose an adaptive database synchronization system that makes distributed data in edge nodes available dynamically to the cloud by balancing between reducing the amount of data that needs to be transmitted and the computational effort needed to do so at the edge. This adapts to the availability of CPU and network resources as well as to the application workload.
An Architecture Proposal for Checkpoint/Restore on Stateful Containers
Rodrigo Herpich Müller, Cristina Meinhardt and Odorico Mendizabal
Containers have been widely adopted for the development of microservices and cloud-native applications. In parts, this popularization stems from the container orchestration platforms support, facilitating the development and continuously integrating largescale applications. However, the established techniques offered by orchestration tools to replicate containers are insufficient to provide high availability and strong consistency for stateful containers. Thus, this paper proposes a Checkpoint/Restore (C/R) service to achieve fault-tolerance on stateful containers. Our service aims to eliminate the checkpoint management burden by adding a transparent C/R service into container orchestration platforms. The checkpointing service, implemented as an interceptor, is responsible for taking snapshots of application containers and coordinating the checkpoint execution with the delivery of requests. In case of faults, a new application container is automatically built from a recent snapshot, and the interceptor resumes the delivery of client requests since that checkpoint. A Proof of Concept using a Kubernetes cluster is implemented and corroborates for the feasibility of the C/R. Some challenges and future directions are discussed.
Karl M. Göschka (Main contact chair)
University of Applied Sciences Technikum Wien
Embedded Systems Institute
A-1200 Vienna, Austria
phone: +43 664 180 6946
goeschka (at) technikum-wien dot at
AT&T Shannon Laboratory
1 AT&T Way, Bedminster, NJ 07921
hiltunen (the at sign goes here) research (dot) att (dot) com
Universidade do Minho
Computer Science Department
Campus de Gualtar
4710-057 Braga, Portugal
phone: +351 253 604 452 / Internal: 4452
rco (at) di dot uminho dot pt
|October 31, 2021 (11:59PM Pacific Time) - extended||Paper submission|
|December 10, 2021||Author notification|
|December 21, 2021||Camera-ready papers|
For general information about SAC, please visit: http://www.sigapp.org/sac/sac2022/
If you have further questions, please do not hesitate to contact us: email@example.com