In this paper, we present RLKube, a Reinforcement Learning (RL)-based custom Kubernetes (K8s) scheduler plugin for optimized task scheduling. RLKube objectives are maximizing resource utilization and Pod throughput as well as improving energy efficiency in a K8s cluster. We used Double Deep Q-Network (DDQN) with Prioritized Experience Replay (PER) and utilized different reward functions to train the RL agent. Also, we have developed corresponding policies for each objective. We have evaluated the effectiveness of RLKube using various datasets simulating a diverse set of realistic load patterns. The results show that RLKube outperforms the default K8s scheduling policies in terms of throughput and energy usage, highlighting its potential for Improving task scheduling in K8s clusters.

Existing data integration frameworks are poorly suited for the special requirements of scientists. To answer a specific research question, often, excerpts of data from different sources need to be integrated. The relevant parts and the set of underlying sources may differ from query to query. The analyses also oftentimes involve frequently changing data and exploratory querying. Additionally, The data sources not only store data in different formats, but also provide inconsistent data access functionality. The classic Extract-Transform-Load (ETL) approach seems too complex and time-consuming and does not fit well with interest and expertise of the scientists. With QUIS (QUery In-Situ), we provide a solution for this problem. QUIS is an open source heterogeneous in-situ data querying system. It utilizes a federated query virtualization approach that is built upon plugged-in adapters. QUIS takes a user query and transforms appropriate portions of it into the corresponding computation model on individual data sources and executes it. It complements the segments of the query that the target data sources can not execute. Hence, it guarantees full syntax and semantic support for its language on all data sources. QUIS's in-situ querying facility almost eliminates the time to prepare the data while maintaining a competitive performance and steady scalability. The present demonstration illustrates interesting features of the system: virtual Schemas, heterogeneous joins, and visual query results. We provide a realistic data processing scenario to examine the system's features. Users can interact with QUIS using its desktop workbench, command line interface, or from any R client including RStudio Server.

This paper presents a secure agent server named J-SAS. J-SAS is a secure and distributed execution environment for mobile agents. The core of the J-SAS is an agent virtual machine fulfilling the same functions as a traditional host machine. J-SAS considers the agent as an autonomous entity with execution code, state, specification and certification parts. J-SAS provides mobility, state dependency, run time code altering, and communications facilities. J-SAS consists of two main parts: agent gateway (AgG) and agent server (AgS). AgG is responsible for receiving agent through HTTP, authentication and delivering it with a passport to AgS. On the other side, AgS gets the agent, checks its passport, translates its code to safe code, initializes its state and runs it in a secure environment. Using resource allocation algorithms, ACL and authentication, J-SAS shows a good defensive mood against unauthorized read/write, denial of service and masquerading attacks.