Since 2015, as part of a Reproducibility Initiative, the International Conference for High Performance Computing, Networking, Storage, and Analysis (SC) has encouraged, and more recently required, all technical papers to include an Artifact Description (AD) Appendix. Each year, the conference selects one accepted paper to form the basis of a Reproducibility Challenge in the Student Cluster Competition (SCC) - a mainstay in the conference program - for the following year. In the SCC, teams of undergraduate students partner with their educational institution and vendors to build, benchmark and operate a small HPC system during the conference week. The Reproducibility Challenge component has them attempt to reproduce parts of the selected work on their own cluster. At SC22, teams worked to reproduce the SC21 paper “Productivity, Portability, Performance: Data-Centric Python” by Ziogas et al. This special section includes a selection of reports from student teams, along with an extended version of that paper, discussing the students’ results...

Scientific workflows enable scientists to undertake analysis on large datasets and perform complex scientific simulations. These workflows are often mapped onto distributed and parallel computational infrastructures to speed up their executions. Prior to its execution, a workflow structure may suffer transformations to accommodate the computing infrastructures, normally involving task clustering and partitioning. However, these transformations may cause workflow imbalance because of the difference between execution task times (runtime imbalance) or because of unconsidered data dependencies that lead to data locality issues (data imbalance). In this article, to mitigate these imbalances, we enhance the workflow lifecycle process in use by introducing a workflow imbalance phase that quantifies workflow imbalance after the transformations. Our technique is based on structural analysis of Petri nets, obtained by model transformation of a data-intensive workflow, and Linear Programming techniques. Our analysis can be used to assist workflow practitioners in finding more efficient ways of transforming and scheduling their workflows. Moreover, based on our analysis, we also propose a technique to mitigate workflow imbalance by data throttling. Our approach is based on autonomic computing principles that determine how data transmission must be throttled throughout workflow jobs. Our autonomic data-throttling approach mainly monitors the execution of the workflow and recompute data-throttling values when certain watchpoints are reached and time derivation is observed. We validate our approach by a formal proof and by simulations along with the Montage workflow. Our findings show that a dynamic data-throttling approach is feasible, does not introduce a significant overhead, and minimizes the usage of input buffers and network bandwidth.

The proliferation of sensors over the last years has generated large amounts of raw data, forming data streams that need to be processed. In many cases, cloud resources are used for such processing, exploiting their flexibility, but these sensor streaming applications often need to support operational and control actions that have real-time and low-latency requirements that go beyond the cost effective and flexible solutions supported by existing cloud frameworks, such as Apache Kafka, Apache Spark Streaming, or Map-Reduce Streams. In this paper, we describe a model-driven and stepwise refinement methodological approach for streaming applications executed over clouds. The central role is assigned to a set of Petri Net models for specifying functional and non-functional requirements. They support model reuse, and a way to combine formal analysis, simulation, and approximate computation of minimal and maximal boundaries of non-functional requirements when the problem is either mathematically or computationally intractable. We show how our proposal can assist developers in their design and implementation decisions from a performance perspective. Our methodology allows to conduct performance analysis: The methodology is intended for all the engineering process stages, and we can (i) analyse how it can be mapped onto cloud resources, and (ii) obtain key performance indicators, including throughput or economic cost, so that developers are assisted in their development tasks and in their decision taking. In order to illustrate our approach, we make use of the pipelined wavefront array.

Summary Recent advances in the type and variety of sensing technologies have led to an extraordinary growth in the volume of data being produced and led to a number of streaming applications that make use of this data. Sensors typically monitor environmental or physical phenomenon at predefined time intervals or triggered by user‐defined events. Understanding how such streaming content (the raw data or events) can be processed within a time threshold remains an important research challenge. We investigate how a cloud‐based computational infrastructure can autonomically respond to such streaming content, offering quality of service guarantees. In particular, we contextualize our approach using an electric vehicles (EVs) charging scenario, where such vehicles need to connect to the electrical grid to charge their batteries. There has been an emerging interest in EV aggregators (primarily intermediate brokers able to estimate aggregate charging demand for a collection of EVs) to coordinate the charging process. We consider predicting EV charging demand as a potential workload with execution time constraints. We assume that an EV aggregator manages a number of geographic areas and a pool of computational resources of a cloud computing cluster to support scheduling of EV charging. The objective is to ensure that there is enough computational capacity to satisfy the requirements for managing EV battery charging requests within specific time constraints.

Recent advances in sensor technologies and instrumentation have led to an extraordinary growth of data sources and streaming applications. A wide variety of devices, from smart phones to dedicated sensors, have the capability of collecting and streaming large amounts of data at unprecedented rates. A number of distinct streaming data models have been proposed. Typical applications for this include smart cites & built environments for instance, where sensor-based infrastructures continue to increase in scale and variety. Understanding how such streaming content can be processed within some time threshold remains a non-trivial and important research topic. We investigate how a cloud-based computational infrastructure can autonomically respond to such streaming content, offering Quality of Service guarantees. We propose an autonomic controller (based on feedback control and queueing theory) to elastically provision virtual machines to meet performance targets associated with a particular data stream. Evaluation is carried out using a federated Cloud-based infrastructure (implemented using CometCloud)-where the allocation of new resources can be based on: (i) differences between sites, i.e., types of resources supported (e.g., GPU versus CPU only), (ii) cost of execution; (iii) failure rate and likely resilience, etc. In particular, we demonstrate how Little's Law-a widely used result in queuing theory-can be adapted to support dynamic control in the context of such resource provisioning.

With an increase in the number of monitoring sensors deployed on physical infrastructures, there is a corresponding increase in data volumes that need to be processed. Data measured or collected by sensors is typically processed at destination or "in-transit" (i.e. from data capture to delivery to a user). When such data are processed in-transit over a shared distributed computing infrastructure, it is useful to provide elastic computational capability which can be adapted based on processing requirements and demand. Where Service Level Agreements (SLAs) have been pre-agreed, such available computational capacity needs to be shared in such a way that any Quality of Service related constraints in such SLAs are not violated. This is particularly challenging for time critical applications and with highly variable and unpredictable rates of data generation (e.g. in Smart Grid applications where energy usage patterns may change unpredictably). Previously, we proposed a Reference net based architectural model for supporting QoS for multiple concurrent data streams being processed (prior to delivery to a user) over a shared infrastructure. In this paper, we describe a practical realisation of this architecture using the Open Nebula Cloud platform. We consider our infrastructure to be composed of a number of nodes, each of which has multiple processing units and data buffers. We utilize the "token bucket" model for regulating, on a per stream basis, the data injection rate into each node. We subsequently demonstrate how a streaming pipeline can be supported and managed using a dynamic control strategy at each node.

With an increasing interest in Electric Vehicles (EVs), it is essential to understand how EV charging could impact demand on the Electricity Grid. Existing approaches used to achieve this make use of a centralised data collection mechanism - which often is agnostic of demand variation in a given geographical area. We present an in-transit data processing architecture that is more efficient and can aggregate a variety of different types of data. A model using Reference nets has been developed and evaluated. Our focus in this paper is primarily to introduce requirements for such an architecture.

Data movement between tasks in scientific workflows has received limited attention compared to task execution. Often the staging of data between tasks is either assumed or the time delay in data transfer is considered to be negligible (compared to task execution). Where data consists of files, such file transfers are accomplished as fast as the network links allow, and once transferred, the files are buffered/stored at their destination. Where a task requires multiple files to execute (from different tasks), it must, however, remain idle until all files are available. Hence, network bandwidth and buffer/storage within a workflow are often not used effectively. We propose an automated workflow structural analysis method for Directed Acyclic Graphs (DAGs) which utilises information from previous workflow executions. The method obtains data-throttling values for the data transfer to enable network bandwidth and buffer/storage capacity to be managed more efficiently. We convert a DAG representation into a Petri net model and analyse the resulting graph using an iterative method to compute data-throttling values. Our approach is demonstrated using the Montage workflow.

The increasing deployment of sensor network infrastructures has led to large volumes of data becoming available, leading to new challenges in storing, processing and transmitting such data. This is especially true when data from multiple sensors is pre-processed prior to delivery to users. Where such data is processed in-transit (i.e. from data capture to delivery to a user) over a shared distributed computing infrastructure, it is necessary to provide some Quality of Service (QoS) guarantees to each user. We propose an architecture for supporting QoS for multiple concurrent scientific workflow data streams being processed (prior to delivery to a user) over a shared infrastructure. We consider such an infrastructure to be composed of a number of nodes, each of which has multiple processing units and data buffers. We utilize the ``token bucket" model for regulating, on a per workflow stream basis, the data injection rate into such a node. We subsequently demonstrate how a streaming pipeline, with intermediate data size variation (inflation/deflation), can be supported and managed using a dynamic control strategy at each node. Such a strategy supports end-to-end QoS with variations in data size between the various nodes involved in the workflow enactment process.

When multiple users execute their streaming applications over a shared Cloud infrastructure, the provider typically captures the Quality of Service (QoS) for each application at a Service Level Agreement (SLA). Such an SLA identifies the cost that a user must pay to achieve the required QoS, and a penalty that must be paid to the user in case the QoS cannot be met. Assuming the maximisation of the revenue as the provider's objective, then it must decide: (i) which user streams to accept for storage and analysis; (ii) how many (computational / storage) resources to allocate to each stream in order to improve overall revenue and minimise cost. In this paper, we analyse revenue models for in-transit streaming applications, executed over a shared Cloud infrastructure under the presence of faulty computational resources. We propose an architecture that features a token bucket process envelop to accept user streams; and a control loop to enable resource allocation, while minimising operational cost.

Abstract Interest in data streaming within scientific workflow has increased significantly over the recent years—mainly due to the emergence of data‐driven applications. Such applications can include data streaming from sensors and data coupling between scientific simulations. To support resource management to enact such streaming‐based workflow, autonomic computing techniques for transmission have been combined with in‐transit processing, so that data elements may be processed in advance, enroute, prior to arrival at the destination. We propose the integration of an autonomic data streaming service (ADSS) with in‐transit processing into a workflow specification. This integration may imply that the associated runtime resource allocation is dependent on environmental conditions and can change for different enactments of the same workflow. In our proposal, our workflow specifications are independent of the constraints imposed by the resource allocation. We express our solutions in terms of Reference nets. We also implement an ADSS utilizing a timed Reference net simulation for predicting future states of the ADSS. There are two advantages: the Reference net which implements the ADSS and the timed model are coincident, and second, token distribution obtained from the Petri net implementation can be utilized to better understand the demand for particular types of resources in the system. Copyright © 2011 John Wiley & Sons, Ltd.

Abstract Scientific workflow systems often operate in highly unreliable, heterogeneous and dynamic environments, and have accordingly incorporated different fault tolerance techniques. We propose an exception‐handling mechanism, based on techniques adopted in programming languages, for modifying at run‐time the structure of a workflow. In contrast to other proposals that achieve the required flexibility by means of the infrastructure, our proposal expresses the exception‐handling mechanism within the workflow language—primarily as two exception‐handling patterns that are exclusively based on the Reference Nets‐within‐Nets formalism (a specific type of Petri nets). When an exception is detected, a workflow in our approach can be re‐written (replaced), based on the particular failure condition that has been detected. This enables workflow users to have better control and understanding of the behaviour of their workflow without having to be aware of the underlying infrastructure. Copyright © 2009 John Wiley & Sons, Ltd.

There is growing interest in promoting interoperability among the many different existent Grid initiatives. The requirements for interlinking of Grids have already been studied from a general point of view. In this paper, we analyse current practice in Grid interactions and show evidence of how Grid interactions may represent a barrier for interlinking of Grids. Our proposal consists of three combined aspects: a) explicitly modelling the interactions (what we call interaction protocols), b) decoupling the execution flow of services from their interactions with other services and c) services have to be able to dynamically select the interaction protocol at the time they have decided to interact with another party. Next, for a clarifying purpose, we show how DVega, an application built on the proposed principles, can use these features as the foundations for interacting with disparate services from different Grids in a flexible and dynamic way.