Workload Consolidation Research Articles

Efficient resource allocation in cloud computing environments using AI-driven predictive analytics

This paper proposes an innovative AI-driven approach for efficient resource allocation in cloud computing environments using predictive analytics. The study addresses the critical challenge of optimizing resource utilization while maintaining high quality of service in dynamic cloud infrastructures. A hybrid predictive model combining XGBoost and LSTM networks is developed to forecast workload patterns across various time horizons. The model leverages historical data from a large-scale cloud environment, encompassing 1000 servers and over 52 million data points. A dynamic resource scaling algorithm is introduced, which integrates the predictive model outputs with real-time system state information to make proactive allocation decisions. The proposed framework incorporates advanced techniques such as workload consolidation, resource oversubscription, and elastic resource pools to maximize utilization efficiency. Experimental results demonstrate significant improvements in key performance indicators, including increasing resource utilization from 65% to 83%, reducing SLA violation rates from 2.5% to 0.8%, and enhancing energy efficiency, with PUE improving from 1.4 to 1.18. Comparative analysis shows that the proposed model outperforms existing prediction accuracy and resource allocation efficiency methods. The study contributes to the field by presenting a comprehensive, AI-driven solution that addresses the complexities of modern cloud environments and paves the way for more intelligent and autonomous cloud resource management systems.

Lavender: An Efficient Resource Partitioning Framework for Large-Scale Job Colocation

Workload consolidation is a widely used approach to enhance resource utilization in modern data centers. However, the concurrent execution of multiple jobs on a shared server introduces contention for essential shared resources such as CPU cores, Last Level Cache, and memory bandwidth. This contention negatively impacts job performance, leading to significant degradation in throughput. To mitigate resource contention, effective resource isolation techniques at the software or hardware level can be employed to partition the shared resources among colocated jobs. However, existing solutions for resource partitioning often assume a limited number of jobs that can be colocated, making them unsuitable for scenarios with a large-scale job colocation due to several critical challenges. In this study, we propose Lavender, a framework specifically designed for addressing large-scale resource partitioning problems. Lavender incorporates several key techniques to tackle the challenges associated with large-scale resource partitioning, ensuring efficiency, adaptivity, and optimality. We conducted comprehensive evaluations of Lavender to validate its performance and analyze the reasons for its advantages. The experimental results demonstrate that Lavender significantly outperforms state-of-the-art baselines. Lavender is publicly available at https://github.com/yanxiaoqi932/OpenSourceLavender.

Multi-resource predictive workload consolidation approach in virtualized environments

The revolution of virtualization technologies and Cloud computing solutions has emphasized the need for energy-efficient and Service level agreement (SLA)-aware resource management techniques in cloud data centers. Workload consolidation in Infrastructure-as-a-Service (IaaS) providers allows for efficient utilization of hardware resources and reduced energy consumption by consolidating workloads onto fewer physical servers. To ensure successful workload consolidation, it is crucial for IaaS providers to carefully estimate the host state and identify overloaded and underloaded hosts, thereby avoiding overly aggressive consolidation. Existing proposals determine the host state depending on its current resource utilization or a single anticipated resource utilization value, and often consider only a single resource type of the host, such as CPU. These limitations may lead to unreliable host state estimations, resulting in excessive and needless service migrations between physical machines (PMs). This, in turn, can lead to extra delays in service execution, degraded performance, increased power consumption, and SLA violations. To address these challenges, we propose a workload consolidation approach that leverages a multi-resource and multi-step resource utilization prediction model. Based on this model, our overload and underload decision-making algorithms consider the forecasted future trend (sequence of future value) of each host resource's utilization, including CPU, memory, and bandwidth. Through extensive experimentations conducted with two real-world datasets, we demonstrate that our approach can significantly reduce power consumption, SLA violation rate, and the number of migrations compared to existing benchmarks.

A novel hybrid meta‐heuristic‐oriented latency sensitive cloud object storage system

SummaryCloud providers must find out how to properly arrange data in a limited count of servers while ensuring latency assurances to reduce total storage expenses. Timeout is also important to consider because it has a substantial impact on response latency. The core aim of this task is to implement a new cloud object storage system strategy that handles challenges like “latency‐sensitive data allocation, latency‐sensitive data re‐allocation, and latency‐sensitive workload consolidation.” The main contribution here is that distributing the latency of the cloud object storage system allows for better data allocation, data reallocation, and workload consolidation. The primary aim is to use the fewest number of servers feasible to fulfill all requests while maintaining their latency requirements, lowering the overall data transmission cost. As a consequence, Whale Butterfly Optimization Method (WBOA) is a novel hybrid meta‐heuristic algorithm that solves NP‐hard problems by combining baseline advanced algorithms. The simulation outcomes reveal that the offered paradigm consistently provides the greatest outcomes regarding throughput utilization, lower latency, higher storage, and number of used nodes when compared to competing techniques.

An Efficient Framework for Utilizing Underloaded Servers in Compute Cloud

In cloud data centers, the consolidation of workload is one of the phases during which the available hosts are allocated tasks. This phenomenon ensures that the least possible number of hosts is used without compromise in meeting the Service Level Agreement (SLA). To consolidate the workloads, the hosts are segregated into three categories: normal hosts, under-loaded hosts, and overloaded hosts based on their utilization. It is to be noted that the identification of an extensively used host or underloaded host is challenging to accomplish. Threshold values were proposed in the literature to detect this scenario. The current study aims to improve the existing methods that choose the underloaded hosts, get rid of Virtual Machines (VMs) from them, and finally place them in some other hosts. The researcher proposes a Host Resource Utilization Aware (HRUAA) Algorithm to detect those underloaded and place its virtual machines on different hosts in a vibrant Cloud environment. The mechanism presented in this study is contrasted with existing mechanisms empirically. The results attained from the study establish that numerous hosts can be shut down, while at the same time, the user's workload requirement can also be met. The proposed method is energy-efficient in workload consolidation, saves cost and time, and leverages active hosts.

Cloud Host Selection using Iterative Particle-Swarm Optimization for Dynamic Container Consolidation

A significant portion of the energy consumption in cloud data centres can be attributed to the inefficient utilization of available resources due to the lack of dynamic resource allocation techniques such as virtual machine migration and workload consolidation strategies to better optimize the utilization of resources. We present a new method for optimizing cloud data centre management by combining virtual machine migration with workload consolidation. Our proposed Energy Efficient Particle Swarm Optimization (EE-PSO) algorithm to improve resource utilization and reduce energy consumption. We carried out experimental evaluations with the Container CloudSim toolkit to demonstrate the effectiveness of the proposed EE-PSO algorithm in terms of energy consumption, quality of service guarantees, the number of newly created VMs, and container migrations.

End-to-End Industrial IoT: Software Optimization and Acceleration

Application of artificial intelligence and machine learning is transforming Industrial Internet of Things (IIoT) segments by enabling higher productivity, better insights, less downtime, and superior product quality. Through AI-inspired innovations, businesses are gaining a sizable competitive edge and product leadership. To realize the promise and potential of AI, software needs to enable fast prototyping and experimentation at scale, and efficiently turn data into valuable insights. Over 100× performance speedup is possible with software that is well parallelized, vectorized, written with better data reuse, has cache blocking, makes good use of prefetching, and above all, is abstracted with familiar industry-standard APIs and libraries used by ML developers and data scientists. As a result of increased software performance efficiency, on the same hardware system, a customer can realize and serve higher insights per second and increase inference throughput. Our software optimizations target a general-purpose CPU, which, in addition to AI models and pipelines, can also be efficient through workload consolidation for other IIoT applications. Both general-purpose workloads and AI model inferencing run on the same hardware targets while minimizing energy, maintenance, and total cost of ownership of a heterogenous infrastructure. Edge IIoT also has unique power and space constraints that can be addressed well with multi-stream inferencing support on general-purpose CPUs. It is imperative to optimize software for all phases of the end-to-end AI pipeline to run “efficient AI” and realize quicker insights, thereby turning them into concrete business results for IIoT solutions. In this article, using optimized frameworks such as Intel Pytorch extensions, Intel Scikit-learn extensions, and Intel distribution of Modin, we get 3.6x-81 × improvement in end-to-end pipeline performance. What this means for the CNN-based anomaly detection pipeline and predictive analytics pipeline solutions is higher throughput of insights and also serving multiple streams of inference across the shop floor, thereby maximizing the potential of the IIoT AI solution. Popular and relevant IIoT AI use cases can realize significant performance improvement when software and AI implementations are purposefully optimized for the target AI hardware and systems.

Holistic Resource Allocation Under Federated Scheduling for Parallel Real-time Tasks

With the technology trend of hardware and workload consolidation for embedded systems and the rapid development of edge computing, there has been increasing interest in supporting parallel real-time tasks to better utilize the multi-core platforms while meeting the stringent real-time constraints. For parallel real-time tasks, the federated scheduling paradigm, which assigns each parallel task a set of dedicated cores, achieves good theoretical bounds by ensuring exclusive use of processing resources to reduce interferences. However, because cores share the last-level cache and memory bandwidth resources, in practice tasks may still interfere with each other despite executing on dedicated cores. Such resource interferences due to concurrent accesses can be even more severe for embedded platforms or edge servers, where the computing power and cache/memory space are limited. To tackle this issue, in this work, we present a holistic resource allocation framework for parallel real-time tasks under federated scheduling. Under our proposed framework, in addition to dedicated cores, each parallel task is also assigned with dedicated cache and memory bandwidth resources. Further, we propose a holistic resource allocation algorithm that well balances the allocation between different resources to achieve good schedulability. Additionally, we provide a full implementation of our framework by extending the federated scheduling system with Intel’s Cache Allocation Technology and MemGuard. Finally, we demonstrate the practicality of our proposed framework via extensive numerical evaluations and empirical experiments using real benchmark programs.

Towards Deadline Guaranteed Cloud Storage Services

More and more organizations move their data and workload to commercial cloud storage systems. However, the multiplexing and sharing of the resources in a cloud storage system present unpredictable data access latency to tenants, which may make online data-intensive applications unable to satisfy their deadline requirements. Thus, it is important for cloud storage systems to provide deadline guaranteed services. In this paper, to meet a current form of service level objective (SLO) that constrains the percentage of each tenant’s data access requests failing to meet its required deadline below a given threshold, we build a mathematical model to derive the upper bound of acceptable request arrival rate on each server. We then propose a Deadline Guaranteed storage service (called DGCloud ) that incorporates three basic algorithms. Its deadline-aware load balancing scheme redirects requests and creates replicas to release the excess load of each server beyond the derived upper bound. Its workload consolidation algorithm tries to maximally reduce servers while still satisfying the SLO to maximize the resource utilization. Its data placement optimization algorithm re-schedules the data placement to minimize the transmission cost of data replication. We further propose three enhancement methods to further improve the performance of DGCloud . A dynamic load balancing method allows an overloaded server to quickly offload its excess workload. A data request queue improvement method sets different priorities to the data responses in a server’s queue so that more requests can satisfy the SLO requirement. A wakeup server selection method selects a sleeping server that stores more popular data to wake up, which allows it to handle more data requests. Our trace-driven experiments in simulation and Amazon EC2 show the superior performance of DGCloud compared with previous methods in terms of deadline guarantees and system resource utilization, and the effectiveness of its individual algorithms.

A power and thermal-aware virtual machine management framework based on machine learning

Energy consumption in data centers grows rapidly in recent years. As a widely-applied energy-efficient method, workload consolidation also has its own limitations that may bring some negative effects, such as performance degradation, QoS violation, localized hotspots and so on, which is especially true when optimal objectives are inherently conflict. In this paper, we present a power and thermal-aware VM management framework called PTM-ML, which relies on machine learning technique to find optimal host configuration based on workload characteristics and cooling system’s working state. Based on such an optimal host configuration, it then makes VM migration and consolidation decisions by enforcing an efficient load-balancing policy, with aiming at achieving a better trade-off between energy efficiency and performance. The prototype of PTM-ML framework is deployed and evaluated in a real-world cloud data center. Extensive experiments are conducted by using different workload traces with distinctive characteristics, and the results are compared with four similar approaches in terms of total energy consumption, real-time power consumption, average latency and etc. Experimental results show that the proposed PTM-ML outperforms the existing approaches in terms of multiple metrics, and it also exhibits better robustness and adaptability in presence of dynamic workloads.

Comparison of workload consolidation algorithms for cloud data centers

AbstractWorkload consolidation is an important method for the efficient operation of cloud data centers, impacting important quality attributes such as resource utilization and power consumption. Many different approaches have been proposed for workload consolidation, but few comparative studies were executed to date. Therefore, it is unclear which of the proposed approaches work best in which situation. In this article, we present a comprehensive simulation‐based comparison of five workload consolidation techniques. We introduce a general framework for workload consolidation techniques to the DISSECT‐CF simulator to foster the development and comparison of efficient data center consolidation algorithms. We use this framework to evaluate the effectiveness of a first fit best fit decreasing heuristic, a custom heuristic, and three population‐based metaheuristics (genetic algorithm, artificial bee colony, and particle swarm optimization). The evaluation is based on a wide variety of real‐world workload traces. The five algorithms are compared in terms of total energy consumption, the duration of the simulation, and the number of migrations. Based on the results, there is no generally best consolidation technique. The results deliver insight into the pros and cons of the algorithms as well as the impact of different parameters. In particular, the results show that population‐based metaheuristics do not offer a significant gain in terms of solution quality to compensate for the increased simulation time.

SDRP: Safe, Efficient, and SLO-Aware Workload Consolidation Through Secure and Dynamic Resource Partitioning

Workload consolidation is a widely-used technique to improve the resource utilization of services computing systems by consolidating latency-critical (LC) and batch workloads on the same physical server. The resource manager for workload consolidation dynamically allocates hardware resources (e.g., cores, caches) to the workloads to maximize the resource utilization while satisfying the service-level objective (SLO) of the LC workloads. Since security-critical hardware resources are dynamically allocated across consolidated workloads, information leakages can be created among workloads through microarchitectural side-channel (SC) attacks. Despite extensive prior works, it is yet to investigate efficient system software support for achieving high resource utilization without compromising the SLO and security of consolidated workloads. To bridge this gap, we propose SDRP, secure and dynamic resource partitioning for safe, efficient, and SLO-aware workload consolidation. As with the state-of-the-art techniques, SDRP dynamically allocates hardware resources to enhance the resource utilization and provide the SLO guarantees. In contrast to the state-of-the-art techniques, SDRP dynamically sanitizes security-critical hardware resources to robustly defeat microarchitectural SC attacks. Our quantitative evaluation demonstrates that SDRP achieves high resource sanitization quality, introduces low performance overheads, delivers high resource utilization with the SLO and security guarantees, and defeats the last-level cache (LLC)-based SC attack.

Experimental and Computational Investigations of the Thermal Environment in a Small Operational Data Center for Potential Energy Efficiency Improvements

Abstract The share of equipment and power use in smaller data centers (DCs) is comparable with that of more massive counterparts. However, they grabbed less attention in the literature despite being less energy-efficient. This study highlights the challenges of setting up a computational fluid dynamics (CFD) model of a 180-m2 small-size high-performance computing (HPC) DC and the validation procedure leading to a reasonably accurate model for the investigation of the thermal environment and potential energy efficiency improvements. Leaky floors, uneven placement of computing equipment and perforated tiles preventing separation of hot and cold air, low-temperature operation, and excessive cooling capacity and fan power were identified sources of energy inefficiency in the DC. Computational fluid dynamics model predictions were gradually improved by using experimental measurements for various boundary conditions (BCs) and detailed geometrical representation of large leakage openings. Eventually, the model led to predictions with an error of less than 1 °C at the rack inlet and less than 5 °C at the rack outlet. The ultimate objective was to use the validated CFD model to test various energy efficiency measures in the form of operational or design changes in line with the best practices. Impact of leakage between the raised floor and the room, reduced airflow rate, cold-aisle and hot-aisle separation, workload consolidation, and higher temperature operation were among the phenomena tested by using the validated CFD model. The estimated power usage effectiveness (PUE) value reduced from 1.95 to 1.40 with the proposed energy efficiency measures.

Coupling energy efficiency and quality for consolidation of cloud workloads

Efficient usage of IT equipment in Data Centers requires modulating their power-consumption according to the actual workload. The most effective strategy to this aim consists in consolidating as many applications as possible on the smallest number of servers, so that idle devices can be shut down or put in low-power states. Usually, this process is driven by computing and networking resources requested by each application (e.g., CPU and RAM) and applies a certain degree of overcommitment, assuming that such resources are not fully used continuously. However, this approach is critical with real workload patterns, which usually change over time; as a matter of fact, consolidation in real scenarios often leads to either low efficiency or violations of Quality of Service (QoS) constraints, depending on the level of overcommitment.In this paper, we investigate a novel consolidation strategy based on an enhanced system model for the Infrastructure-as-a-Service cloud paradigm, which targets a better trade-off between Energy Efficiency and Quality of Service. We explicitly target modular cloud applications, which design is split into multiple components deployed in Virtual Machine (VM)s or containers. Our consolidation strategy allows to “freeze” parts of the application which are not currently used, making them available when requested with minimal latency. This improves energy saving with respect to other approaches, especially when idle VMs are present for backup or redundancy purposes, without degrading the service level. We compare multiple heuristics available in Optaplanner to solve our consolidation problem, and investigate improvements with respect to a more traditional approach. Our evaluation includes both simulations and experimentation in a real test-bed. The comparison shows that the Late Acceptance algorithm on average finds better solutions than other alternatives and energy efficiency improves up to 40% with respect to more conventional strategies, with deterioration of QoS indexes below 1%.

Energy-efficient strategy for virtual machine consolidation in cloud environment

An important issue of energy efficiency in cloud environment is to perform more jobs while consuming less amount of power. Virtual machine consolidation remains the most deployed strategy to manage both performance and energy consumption. Most of existing energy efficiency techniques save energy against the cost on performance degradation. Consolidation techniques leverage thresholds to detect overloaded and underloaded hosts that could be vacated to achieve optimal balance between host utilization and energy consumption. In this research, we propose an energy-efficient strategy (EES) to consolidate virtual machines in cloud environment with an aim of reducing the energy consumption while completing more tasks with the highest throughput. Our proposal makes use of the performance-to-power ratio to set upper thresholds for overload detection. In addition, EES considers the overall data center workload utilization to set lower thresholds, which can reduce the number of virtual machine migrations. The simulation results show that EES leads to energy-efficient workload consolidation with the minimal number of migrations and less energy consumption. The results conclude that EES saves energy consumption without compromising user’s workload requirement.

Interactive Context for Mobile OS Resource Management

This paper presents an application-transparent execution context for mobile operating systems that reflects the criticality of current execution on the user interactivity. We track Android / Linux system-level events including semantic syscalls, UI actions, and standard inter-process and intra-process communication interfaces that signal the initiation and propagation of interactivity-related executions. This interactive context enables new optimizations in CPU scheduling and power state management. We recognize that online resource scheduling is susceptible to inaccurate contexts—e.g., misidentifying an interactive execution as background leads to possible priority inversion by schedulers. In particular, low-level OS events such as kernel block / wake-up carry ambiguous semantics that could induce false context propagations. We selectively expose such uncertainties to the schedulers so that executions with uncertain contexts are prioritized behind interactive executions but otherwise run as fast as possible to minimize the priority inversion. We further recognize the importance of prompt propagation of the interactivity context at the earliest possible moment to facilitate immediate prioritization. We utilize the interactivity context to enable differential per-core frequency control and background workload consolidation. Experiments on a quad-core smartphone with a dozen mobile workload scenarios show that we can, on average, save 13 percent energy with only 1 percent degradation on the interactive performance.

Multi-Tier Workload Consolidations in the Cloud: Profiling, Modeling and Optimization

Reducing tail latency becomes increasingly important to improve the user-perceived service experience. User-facing latency-sensitive cloud applications typically contain multiple interactive tiers (e.g., Web, App, Database) running in different virtual machines (VMs) with complex interaction patterns. However, such interactions between VMs in different tiers are often neglected in previous VM consolidation methods, resulting in poor application performance. In this article, we study the consolidation of multi-tier interactive workloads from a new perspective of user-perceived tail latency. We propose a novel profiling-based consolidation methodology to satisfy tail latency requirements while reducing the number of used physical machines. To achieve such a goal, we first perform large-scale profiling experiments under various consolidation settings in a KVM virtualized private cluster to establish the empirical performance values. We consider two key factors that affect the tail latency of multi-tier workloads: <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">interference</i> with co-located VMs and <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">interaction</i> between tiers. We model the consolidation of multi-tier workloads as an optimization problem with different objectives and constraints, and derive the consolidation schedule. We implement and evaluate the proposed models, as well as comparing with other methods (i.e., <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">without</i> profiling or <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">without</i> considering interaction influence). Extensive experimental results show that the proposed method is able to reduce up to <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">5X</i> tail latency, compared with the method <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">without</i> profiling and up to <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1.3X</i> tail latency, compared with the method <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">without</i> considering the interaction influence between different tiers.

Micro-Economic Benefits of Peer-Producing Containerized Network Functions

Discrete, non-virtualized network elements are characterized by large costs, limited functions, vendor lock-in, and limited orchestration. Virtualization technologies like virtual machines (VM) and containers have expanded the scope of virtual resource utilization through consolidation of workloads that were previously running on multiple servers by running them on a single server. With the advent of Network Functions Virtualization (NFV), industries are able to reduce the micro-economic factors associated with vendor proprietary model such as transaction costs and (physical and human) asset specificity to deal with vendor vulnerabilities in contractual relationships because Virtual Network Functions (VNFs) can virtualize dedicated networking functions that were traditionally performed by vendor appliances such as routers, switches, firewalls, and load balancers. Even though virtualization technologies (VMs and containers) and NFV have demonstrated their benefits in the market, little attention has been devoted to the development and adoption of containers to build VNFs. This research paper identifies micro-economic factors, such as transaction costs, associated with searching, buying, provisioning, and maintenance of vendor proprietary appliances and compares them with the coordination costs associated with the adoption of containerized VNFs. This comparative analysis could be used to identify the type of network operators that could serve as key organizers (the network operator who can benefit largely by adopting containerized VNFs) of an open source peer production model as well as other firms that could serve as individual contributors. Furthermore, to identify various rewards and incentives that a managerial firm can leverage to motivate its employees to participate in such an effort, a quantitative survey was conducted (with Tier 1, Tier 2 and Tier 3 Service Providers) to identify the managerial incentives such as bonuses, rewards, peer recognition, and promotion targeting varied network operator firms to accurately capture and analyze employee interests/motivation. Finally, the paper shows a measuring framework to evaluate individual contributors based on the project modularity and indicates the viability of this model.

Utilization Driven Model for Server Consolidation in Cloud Data Centers

The application of cloud computing has diversified with the adoption of Internet of Things (IoTs) and edge computing. However, it has increased the uncertainty of workload demand; thus, the efficient utilization of cloud computing resources become more challenging. Traditionally, dynamic consolidation of workload inside cloud data centers relies on identifying overload and under-load hosts using either static or dynamic threshold value. In this paper, we propose a Utilization Driven Model (UDM) model to estimate the number of under-utilized and over-utilized processing machines through percentile ranks of low and high utilization from mean value of resource utilization of hosts and the value of mean absolute deviation of resource demand. UDM swiftly reacts to any change in workload demand and adapts the system to the current demand of resource utilization. The UDM approach not only impacts the energy consumption and quality of service but also increases the elastic nature of cloud by robustly managing the sudden changes in workload. Experiment results show that UDM is an efficient server consolidation technique, improving 30% energy, 40% quality of service compared to contemporary techniques. Thus, the UDM is more robust to support stochastic resource demand compared to traditional techniques.

Cacol: A zero overhead and non-intrusive double caching mitigation system

In a virtualized server, a page cache (which caches I/O data) is managed in both the hypervisor and the guest Operating System (OS). This leads to a well-known issue called double caching. Double caching is the presence of the same data in both the hypervisor page cache and guest OS page caches. Some experiments we performed show that almost 64% of the hypervisor page cache content is also present in guest page caches.Therefore, double caching is a huge source of memory waste in virtualized servers, particularly when I/O disk-intensive workloads are executed (which is common in today’s data centers). Knowing that memory is the limiting resource for workload consolidation, double caching is a critical issue.This paper presents a novel solution (called Cacol) to avoid double caching. Unlike existing solutions, Cacol is not intrusive as it does not require guest OS modification and induces very little performance overhead for user applications. We implemented Cacol in KVM hypervisor, a very popular virtualization system. We evaluated Cacol and compared it with Linux default solutions. The results confirm the advantages (no guest modification and no overhead) of Cacol against existing solutions.