Combining different heterogeneous components into a full HPC system results in combinatorial effects in their complexity.It is a huge challenge to design systems such that they can be used efficiently by the expected workloads, particularly, when the workload is very heterogeneous.Modular systems can help, deciding according to the user portfolio how much weight a particular module should get, and what connectivity is required within and between modules.To deal with these challenges, integrated projects that cover all levels of the HPC ecosystem are needed.Also, interoperability and exchangeability of components, both hardware and software, should be easier to give system designers and users, alike, more flexibility.This special issue calls for recent research which focused on the heterogeneous HPC for IDA, such as memory management, workload management for heterogeneous systems and so as the heterogeneity in storage technologies.

Summary Teaching‐learning‐based optimization (TLBO) algorithm is a population‐based meta‐heuristic algorithm that was created to solve single‐objective optimization problems. The teaching‐learning mechanism of a classroom inspires it. TLBO suffers from weak exploration. As a result, its performance is not good for solving multimodal problems. To turn TLBO into a tool for solving multimodal problems and maintaining good diversity, we made significant modifications into the learning process of the fundamental TLBO. The proposed algorithm produces more diverse solutions and works better for solving multimodal problems. This newly created variant of TLBO is called “Intelligent‐Teaching‐Learning‐Based Optimization (I‐TLBO) algorithm.” I‐TLBO's performance is evaluated against the most recent standard benchmark function, CEC‐06, 2019, and it is discovered that I‐TLBO outperforms the other algorithms. After that, I‐TLBO was applied for flowtime‐aware‐cost minimization of the workflow executions in cloud datacenter. To solve these scheduling problems, I‐TLBO and other metaheuristic algorithms are simulated in CloudSim and tested over scientific workflows such as Inspiral, Montage, SIPHT, sample, Cybershake, and Epigenomics workflows. Finally, it is found that I‐TLBO reduces flowtime and cost both by 28.48%, 11.30%, 17.64%, 13.22%, 11.45%, and 14.71% in comparison to the second best performing algorithm while executing the standard workflow in cloud.

With the rapid advancements in the field of autonomous driving, the need for faster and more accurate object detection frameworks has become a necessity. Many recent deep learning-based object detectors have shown compelling performance for the detection of large objects in a variety of real-time driving applications. However, the detection of small objects such as traffic signs and traffic lights is a challenging task owing to the complex nature of such objects. Additionally, the complexity present in a few images due to the existence of foreground/background imbalance and perspective distortion caused by adverse weather and low-lighting conditions further makes it difficult to detect small objects accurately. In this letter, we investigate how an existing object detector can be adjusted to address specific tasks and how these modifications can impact the detection of small objects. To achieve this, we explore and introduce architectural changes to the popular YOLOv5 model to improve its performance in the detection of small objects without sacrificing the detection accuracy of large objects, particularly in autonomous driving. We will show that our modifications barely increase the computational complexity but significantly improve the detection accuracy and speed. Compared to the conventional YOLOv5, the proposed iS-YOLOv5 model increases the mean Average Precision (mAP) by 3.35% on the BDD100K dataset. Nevertheless, our proposed model improves the detection speed by 2.57 frames per second (FPS) compared to the YOLOv5 model.

Abstract Workflow scheduling is an important way to manage the execution of a workflow. It introduces the concept of providing suitable resources to workflow tasks in order to finish workflow execution and meet the user's objectives. However, the problem becomes more complex when scheduling must balance two conflicting objectives, such as minimizing execution cost and maximizing load across all computing resources. A workflow has many interdependent tasks, and the cloud datacenter has many computing resources to execute the workflow. There can be an asymptotically infinite number of mappings of tasks‐to‐computing resources. Every mapping produces different execution costs with different workloads on computing resources. The main challenge for the researcher is to develop an intelligent scheduling algorithm to identify an optimal mapping that produces minimal execution cost with fair workload distribution on resources. We developed a novel meta‐heuristic algorithm named Investment‐Based Optimization (IBO) to identify an optimal mapping. The IBO algorithm was first tested on optimization benchmark functions and then simulated in CloudSim to see its performance for scheduling workflows. Finally, IBO was tested over Montage, Epigenomics, Sipht, and a sample workflow, and it was found that IBO reduces execution costs by 33%, 16%, 16.36%, and 20% with a fair workload distribution.

Abstract Teaching learning‐based optimization (TLBO) was proposed by Rao to solve optimization problems. It is based on the theory of teaching‐learning mechanism. Although it performs well in unimodal problems yet its performance is not good in multimodal problems. To further improve this algorithm's performance and make it suitable for both unimodal problems and multimodal problems, we made some major changes in the theory and the algorithm's operators. The proposed algorithm is able to capture diverse optimal solutions in less number of iterations and is very good for solving multimodal problems. This newly created variant of TLBO is named generalized TLBO (GTLBO). The performance of GTLBO is tested on CEC−06, 2019 benchmark functions and other 15 classical benchmark functions, and it is found that the proposed algorithm is performing better comparatively. Then it is simulated for solving the workflow scheduling problem in CloudSim. Standard scientific workflow applications as Montage, Epigenomics, Sipht, and a sample workflow are used as dataset to test algorithms' performance in cloud environments. Our proposed approach, GTLBO, provides the proper distribution of workloads and offers minimal execution‐cost for the workflow applications. Results reflect the supremacy of the proposed algorithm GTLBO comparatively.

Summary A time‐limited data access control scheme allows a user's access to the data files only for a specified time period. A cryptographic solution to the time‐limited access control problem is by encrypting each data group associated with a time period with a distinct key. The data is encrypted by the data owner. The respective decryption keys are then distributed to authorized users by the data owner. A user requires one secret decryption key storage for each authorized time period. To reduce the secret key storage with each user, time‐limited hierarchical key management schemes are generally used. Many such schemes are proposed in the recent years. The objective of these schemes is system efficiency and data security. Construction of such schemes become more challenging when data is outsourced to an untrusted third party service provider. In current work, an efficient and secure time‐limited hierarchical key assignment scheme is proposed for key management suitable for data outsourcing scenario. We compare it with the other recent similar schemes. The scheme is formally proved against the modern stronger security notion called key indistinguishability.

In this paper, a variant of particle swarm optimization (PSO) is presented to handle the problem of stagnation encounters in PSO which may lead to get it trapped in local optima and premature convergence particularly in multimodal problems. The proposed scheme Dynamic-PSO (DPSO) does not disturb the fast convergence characteristics of PSO by keeping the basic concept of PSO unaffected. When particles personal best and swarm's global best position do not improve in successive generation i.e. start stagnating DPSO provides dynamicity to particles externally in such a manner that stagnated particles move towards potentially better unexplored region to maintain diversity as this increases chance to recover from stagnation. By identifying and curing stagnated particles, it also avoids the problems of getting trapped in local optima and premature convergence. We have compared the proposed algorithm DPSO with basic PSO and its widely accepted variants over 24 benchmark functions provided by Black-Box Optimization Benchmarking (BBOB 2013). Results show that the proposed variant performs better in comparison with other peer algorithms.

An Environmental adaption Method (EAM) has been established earlier [2]. In this paper an Environmental Adaption Method for Dynamic Environment (EAMD) has been proposed, which has been specially designed with real valued parameters in dynamic environment. It simulates an environment which gradually becomes more deadly for its inhabitants and only the individuals who are able to adapt to this changing environment will survive and improve their fitness over time. This change in the environment causes the solutions to converge towards the optimal solutions. EAMD is compared with two cellular genetic algorithms (grid16, grid100), a single population genetic algorithm (ga100) and a hill climber on the Black Box Optimization test-bed at dimensions 2D and 10D on a set of 24 benchmark functions. The proposed algorithm gives better results than the existing algorithms.

Last decade has noticed an enormous growth of Internet based information services and applications such as Email, Teleconferencing, Videophony or VoIP etc. All these applications have different QoS (Quality of Service) expectations (bandwidth, delay, jitter and reliability etc). The purpose and the benefits from these applications could be blemished if the underlying communication network does not fulfills the QoS requirements. However, different applications have different prioritized QoS requirements. Some applications put heavy demands of bandwidth, some require a more degree of reliability while some expect a negligible response time and so on. So there is always a need of routing strategy which is not only efficient and precise but guarantees the QoS measure as according to application's priorities also. Tackling the issue, this paper presents the use of priority based Evolutionary Multi-objective Optimization algorithms to find the optimal routes for the data flows of various QoS classes via optimizing multiple QoS parameters namely response time, bandwidth requirements and reliability according to their importance for the application under consideration.

In this paper we will study one multi objective optimization problem, which is related to small-scale foundry. Rotary furnace is used in small-scale foundry to melt the metal. To increase the production of a foundry we have to increase melting rate of the rotary furnace. We will use NSGA-111 to maximize the melting rate of rotary furnace by minimizing the amount of fuel used.