ABSTRACT In today's distributed computing environments, the rapid generation of large‐scale data from diverse sources poses significant challenges in terms of storage, management, and processing, particularly for traditional relational databases. Hadoop has emerged as a widely adopted framework for handling such data through parallel processing across distributed clusters. Despite its advantages in scalability, flexibility, and fault tolerance, Hadoop suffers from inefficiencies related to high data access latency, redundant computations, and I/O overhead, which degrade overall system performance. To mitigate these issues, researchers have proposed various caching mechanisms aimed at improving data access time, enhancing data locality, minimizing duplicate computations, and optimizing resource utilization. This paper provides a comprehensive survey and novel classification of existing caching strategies for Hadoop, categorizing them based on the specific Hadoop performance bottlenecks they address. A detailed comparative analysis is provided based on critical caching characteristics such as cached item type, cache management policies, replacement strategies, and access patterns. To assess the effectiveness of these caching mechanisms, their impact on key Hadoop performance metrics is evaluated. Also, statistical insights are presented, highlighting the percentage of reviewed studies addressing specific Hadoop performance challenges and the frequency of performance metrics used for evaluation. Finally, this survey identifies hybrid caching as a promising future trend and proposes a novel approach termed Hybrid Intelligent Cache (HIC) as an example. HIC combines the strengths of two previously developed methods from distinct categories. The first method is the Hybrid Support Vector Machine–Least Recently Used (H‐SVM‐LRU) algorithm, which enhances the traditional LRU cache replacement strategy by employing a Support Vector Machine (SVM) to predict future data access patterns for intelligent eviction. The second is Cache Locality with Q‐Learning in MapReduce Scheduling (CLQLMRS), a reinforcement learning–based scheduling technique that optimizes task allocation by maximizing both cache locality and data locality. Experimental results demonstrate that HIC yields an average 31.2% improvement in job execution time, marking a significant advancement in intelligent caching for Hadoop ecosystems.

With the current high levels of energy consumption of data centers, reducing power consumption by even a small percentage is beneficial. We propose a framework for thermal-aware workload distribution in a data center to reduce cooling power consumption. The framework includes linearization of the general optimization problem and proposing a heuristic to approximate the solution for the resulting Mixed Integer Linear Programming (MILP) problems. We first define a general nonlinear power optimization problem including several cooling parameters, heat recirculation effects, and constraints on server temperatures. We propose to study a linearized version of the problem, which is easier to analyze. As an energy saving scenario and as a proof of concept for our approach, we also consider the possibility that the red-line temperature for idle servers is higher than that for busy servers. For the resulting MILP problem, we propose a heuristic for intelligent rounding of the fractional solution. Through numerical simulations, we compare our heuristics with several existing algorithms. In addition, we evaluate the performance of the solution of the linearized system on the original system. Finally, the results show that the proposed approach can reduce the cooling power consumption by more than 10 percent compared to the case of continuous utilizations and a single red-line temperature.Note to Practitioners—We present a holistic approach for thermal-aware workload distribution for power consumption reduction in data centers. We suggest that when thermal and power consumption models can be linearized, a model-independent approach can be used for optimization purposes. The standard linear problem that results presents some technical challenges to solve, for which we present intuitive and effective solution heuristics. The heuristics are simple enough that they could be used for real-time calculations. The result is that customized models and problems can be avoided (a linear model could be directly constructed from operational data, if available), allowing for the simplification of operational control problems. Our approach is evaluated for a high-fidelity model of a real data center, where both the linearization and optimization components are validated. Finally, we show how this approach can be used to effectively solve the operational problem of workload distribution in the presence of utilization-dependent server red-line temperatures.

In the era of cloud computing, accessing virtual computing resources has become increasingly convenient for users to meet their demands. Cloud providers offer two primary payment plans for virtual resource provisioning: reservation and on-demand. The reservation plan requires users to reserve resources and pay upfront, making it more cost-effective for long-term requirements despite demand uncertainty. Conversely, the on-demand plan charges based on actual resource usage, making it more expensive but suitable for short-term needs. Efficient resource provisioning is crucial to balance user demands and costs, as inefficient provisioning can lead to high costs. A key challenge is determining the optimal number of resources to reserve to accommodate uncertain demands while minimizing costs. This paper addresses the resource reservation problem in cloud environments by focusing on the optimal reservation of virtual machines (VMs). We propose a linear regression approach that fits a linear function to features such as past demands and previously reserved instances that are still available to determine the quantity of VMs to reserve. Our model assigns specific weights to these features to predict required reserved instances, minimizing the overall cost, including both the expense of reserving resources and renting additional on-demand resources as needed. Our evaluation, based on real standard workload traces, demonstrates the effectiveness of our approach in achieving cost-efficient resource provisioning, reducing the total cost by efficiently balancing reserved and on-demand resources.

The Long Short-Term Memory (LSTM) architecture is one of the most successful types of Recurrent Neural Networks (RNNs). However, the number of parameters that LSTMs need to achieve acceptable performance might be larger than desired for standard devices. In this work, an Extended LSTM (E-LSTM) architecture is proposed to reduce the number of parameters needed to achieve similar performance to LSTMs. The architecture of the proposed E-LSTM is characterized by higher and explicit connectivity between distant past and current cell states' values, increasing the likelihood of a higher information transmission between data points separated by long lags. Analysis of possible nonlinear relations in the data sets aids in the process of determining an appropriate placement of increased connectivity, performed using the Distance Correlation (DC) method. The proposed E-LSTM reduces the number of parameters needed, by an order of magnitude for some cases, with acceptable increases in both CPU time and memory needed for the training process.

Energy consumption of today's datacenters is a constant concern from the standpoints of monetary and environmental costs. We model a datacenter as a queueing system, where each server can be switched on or off, with the time to switch a server on being nonnegligible. Previously derived structural properties of the optimal policy allow us to intelligently select policies to analyse further. Using the recursive renewal reward technique, we offer an exact analysis of these policies alongside offering insights, observations, and implications for how these systems behave. In particular, we provide insight into the question of the number of servers that should remain on at all times under a general cost function.

Dynamic scheduling problems consist of both challenging combinatorics, as found in classical scheduling problems, and stochastics due to uncertainty about the arrival times, resource requirements, and processing times of jobs. To address these two challenges, we investigate the integration of queueing theory and scheduling. The former reasons about long-run stochastic system characteristics, whereas the latter typically deals with short-term combinatorics. We investigate two simple problems to isolate the core differences and potential synergies between the two approaches: a two-machine dynamic flowshop and a flexible queueing network. We show for the first time that stability, a fundamental characteristic in queueing theory, can be applied to approaches that periodically solve combinatorial scheduling problems. We empirically demonstrate that for a dynamic flowshop, the use of combinatorial reasoning has little impact on schedule quality beyond queueing approaches. In contrast, for the more complicated flexible queueing network, a novel algorithm that combines long-term guidance from queueing theory with short-term combinatorial decision making outperforms all other tested approaches. To our knowledge, this is the first time that such a hybrid of queueing theory and scheduling techniques has been proposed and evaluated.

As energy costs and energy used by server farms increase, so does the desire to implement energy-aware policies. Although under some metrics, optimal policies for single as well as multiple server systems are known, a number of metrics remain without sufficient knowledge of corresponding optimal policies. We describe and analyse a model to determine an optimal policy for on/off single server systems under a broad range of metrics that are based on expected response time, energy usage, and switching costs. We leverage this model in the determination of routing probabilities to show a range of nontrivial optimal routing probabilities and server configurations when energy concerns are a factor.

Advances in technology and economical pressure have forced many organizations to consider the migration of their legacy systems to newer platforms.

Desktop Grids are rapidly gaining popularity as a cost-effective computing platform for the execution of applications with extensive computing needs. As opposed to grids and clusters, these systems are characterized by having a non-dedicated infrastructure. These unique characteristics need to be considered in developing resource management strategies for Desktop Grids. Several frameworks for the performance evaluation of resource management strategies have been suggested for grids. However, similar projects for Desktop Grids are still lacking. This paper presents MGST, the first performance testing framework for Desktop Grids. We discuss the design of the tool and show how it can be used to analyze and improve the performance of an existing Desktop Grid scheduling policy.

Privacy and security are critical requirements for using patient profiles in distributed healthcare environments. The amalgamation of new information technology with traditional healthcare workflows for sharing patient profiles has made the entire system vulnerable to security and privacy breaches. In this paper we present a novel access control model based on a framework designed for data and service interoperability in the healthcare domain. The proposed model for customizable access control captures the dynamic behavior of the user and determines access rights accordingly. The model is generic and flexible in the sense that an access control engine dynamically receives security effective factors from the subject user, and identifies the privilege level in accessing clinical data using different specialized components within the engine. Standard data representation formats are used to make the model compatible with different healthcare environments. The access control engine uses a flow-based approach to follow the user's behavior. The proposed model is supported by a real world case study.

Resource management systems (RMS) are an important component in heterogeneous computing (HC) systems. One of the jobs of an RMS is the mapping of arriving tasks onto the machines of the HC system. Many different mapping heuristics have been proposed in recent years. However, most of these heuristics suffer from several limitations. One of these limitations is the performance degradation that results from using outdated global information about the status of all machines in the HC system. This paper proposes several heuristics which address this limitation by only requiring partial information in making the mapping decisions. These heuristics utilize the solution to a linear programming (LP) problem which maximizes the system capacity. Simulation results show that our heuristics perform very competitively while requiring dramatically less information.

We consider a system with a dispatcher and several identical servers in parallel. Task processing times are known upon arrival. We first study the impact of the local scheduling policy at a server. To this end, we study random routing followed by a priority scheme at each server. Our numerical results show that the performance (mean waiting time) of such a policy could be significantly better than the best known suggested policies that use FCFS at each server. We then propose to use multi-layered round robin routing, which is shown to further improve system performance. Our analysis involves a combination of comparing analytic models, heavy traffic asymptotic and numerical work.