Machine Learning (ML) training is a growing workload in high-performance computing clusters and data centers; furthermore, it is computationally intensive and requires substantial amounts of energy with associated emissions. To the best of our knowledge, previous works in the area of load management have never focused on decreasing the carbon emission of ML training workloads. In this paper, we explore the potential emission reduction achievable by leveraging the iterative nature of the training process as well as the variability of CO2signal intensity as coming from the power grid. To this end, we introduce two emission-aware mechanisms to shift the training jobs in time and migrate them between geographical locations. We present experimental results on power and carbon emission of the training process together with delay overheads associated with emission reduction mechanisms, for various, representative, deep neural network models. The results show that following emission signals, one can effectively reduce emissions by an amount that varies from 13% to 57% of the baseline cases. Moreover, the experimental results show that the total delay overhead for applying emission-aware mechanisms multiple times is negligible compared to the jobs’ completion time.

The existence of different process variants is inevitable in many modern organizations. However, variability in business process support has proven to be a challenge as it requires a flexible business process specification that supports the required process variants, while at the same time being compliant with policies and regulations. Declarative approaches could support variability, by providing rules constraining process behavior and thereby allowing different variants. However, manual specification of these rules is complicated and error-prone. As such, tools are required to ensure that duplication and overlap of rules is avoided as much as possible, while retaining maintainability. In this paper, we present an approach to represent different process variants in a single compound prime event structure, and provide a method to subsequently derive variability rules from this compound prime event structure. The approach is evaluated by conducting an exploratory evaluation on different sets of real-life business process variants, including a real-life case from the Dutch eGovernment, to demonstrate the effectiveness and applicability of the approach.

The concept of software as a service has been increasingly adopted to develop distributed applications. Ensuring the reliability of loosely coupled compositions is a challenging task because of the open, dynamic, and independent nature of composable services; this is especially true when the execution of a service-based process relies on independent but correlated services. Transactions are the prototypical case of compositions spanning across multiple services and needing properties to be valid throughout the whole execution. Although transaction protocols and service composition languages have been proposed in the past decade, a true viable and effective solution is still missing. In this article, we propose a systematic aspect-based approach to integrating transactions into service compositions, taking into account the well-known protocols: Web Service Transaction and Business Process Execution Language (BPEL). In our approach, transaction policies are first defined as a set of aspects. They are then converted to standard BPEL elements. Finally, these transaction-related elements and the original BPEL process are weaved together, resulting in a transactional executable BPEL process. At runtime, transaction management is the responsibility of a middleware, which implements the coordination framework and transaction protocols followed by the transactional BPEL process and transaction-aware Web services. To automate the proposed approach, we developed a supporting platform called Salan to aid the tasks of defining, validating, and weaving aspect-based transaction policies, and of deploying the transactional BPEL processes. By means of a case study, we demonstrate the proposed approach and evaluate the performance of the supporting platform. Experimental results show that this approach is effective in producing reliable business processes while reducing the need for direct human involvement.

Domotics, concerned with the realization of intelligent home environments, is a novel field which can highly benefit from solutions inspired by service-oriented principles to enhance the convenience and security of modern home residents. In this work, we present an architecture for a smart home, starting from the lower device interconnectivity level up to the higher application layers that undertake the load of complex functionalities and provide a number of services to end-users. We claim that in order for smart homes to exhibit a genuinely intelligent behavior, the ability to compute compositions of individual devices automatically and dynamically is paramount. To this end, we incorporate into the architecture a composition component that employs artificial intelligence domain-independent planning to generate compositions at runtime, in a constantly evolving environment. We have implemented a fully working prototype that realizes such an architecture, and have evaluated it both in terms of performance as well as from the end-user point of view. The results of the evaluation show that the service-oriented architectural design and the support for dynamic compositions is quite efficient from the technical point of view, and that the system succeeds in satisfying the expectations and objectives of the users.

Business processes that span organizational borders describe the interaction between multiple parties working towards a common objective. They also express business rules that govern the behavior of the process and account for expressing changes reflecting new business objectives and new market situations. We developed a service request language and support framework that allow users to formulate their requests against standard business processes.19 In this paper, we extend the approach by presenting a framework capable of automatically associating business rules with relevant processes involved in a user request. This framework plans and monitors the execution of the request and assertions against services underlying these processes. Definitions and classifications of business rules (named assertions in the paper) are given together with an assertion language for expressing them. The framework is able to handle the non-determinism typical for service-oriented computing environments and it is based on the interleaving of planning and execution. Interestingly, the language is able to express both functional and non-functional aspects of the assertions.