CPU Workload Research Articles

NLTSP: A cost model for tensor program tuning using nested loop trees

This paper introduces NLTSP, a deep learning-based cost model designed to optimize tensor program performance in deep learning compilers. NLTSP, short for Nested Loop Tree Structure Processing, facilitates tensor program tuning by extracting information directly from the nested loop tree structure of sampled programs. NLTSP extracts features upstream in the compilation flow and eliminates the need for complex feature engineering. By utilizing a unified format for CPU and GPU architectures and extracting simple high-level features, NLTSP significantly accelerates feature extraction speed while maintaining performance accuracy. We have integrated this technology into Ansor, a leading search framework in the TVM compiler, and conducted experiments. Compared with TenSet MLP, the state-of-the-art cost model utilizing Ansor features as inputs, NLTSP achieves feature extraction speeds on CPU and GPU that are, on average, 97.9 times and 41.4 times faster, respectively, and can reduce the average search time for CPU and GPU workloads by 2.50 times and 4.11 times, respectively. It is worth noting that NLTSP is not specifically designed for Ansor. Any auto-tuning framework capable of representing scheduled tensor programs as nested loop trees can potentially benefit from using NLTSP to achieve superior performance. The code is available at https://github.com/xhq0/NLTSP.

EPO‐R: An efficient garbage collection scheme for long‐term transactions

SummaryThis article proposes EPO‐R, which is an efficient garbage collection scheme designed for multi‐version concurrency control (MVCC) protocols. MVCC generates a version for each update operation, and it can exhaust physical memory for dynamically changing environments such as IoT payments. Eager pruning of obsolete versions (EPO) is a novel garbage collection technique for long‐term transactions. We found room for improvement in EPO, which triggers reclamation invocation. EPO is triggered for each write operation, and it wastes CPU resources. The proposed method EPO‐R is triggered on a read operation to address this issue. The result of experiments with 64 CPU cores and workloads with the combination of short‐term and long‐term transactions demonstrated that EPO‐R exhibited 6.3 times higher throughput than that of EPO. The reason for this can be explained by analyzing the number of unreclaimable versions.

CIPO: Efficient, lightweight and programmable packet scheduling

With the rapid development of network devices and increasingly high CPU workloads, packet scheduling will have to be offloaded to hardware. Programmable packet scheduling allows scheduling algorithms to be programmed into network devices without modifying the hardware. It not only retains the flexibility of software but also the scalability of hardware. In existing primitives, the most expressive Push-In-ExtractOut (PIEO) is prohibitively expensive to implement due to its complexity. While its variant, Push-In-Pick-Out (PIPO), offers some improvements, it suffers from insufficient scalability. In this paper, we propose the Classify-In-Push-Out (CIPO) primitive. The core idea of CIPO is to track the rank and predicate of recent packets through a sliding window, filter and classify packets using a prediction-based two-dimensional classification algorithm and a finite number of First-In-First-Out (FIFO) queues. Through theoretical analysis and evaluation with a range of real workloads, CIPO proves that it has a scheduling performance similar to the most expressive scheduling primitive. Importantly, CIPO requires fewer hardware resources while still providing sufficient expressiveness. Primitive on FPGA show that the CIPO-based scheduler achieves an average of 1.24× higher throughput than the PIEO-based scheduler but uses only an average of 26.2% of look-up tables (LUTs) and 12.2% of the block RAM of the latter.

Reduced CPU Workload for Human Pose Detection with the Aid of a Low-Resolution Infrared Array Sensor on Embedded Systems

Modern embedded systems have achieved relatively high processing power. They can be used for edge computing and computer vision, where data are collected and processed locally, without the need for network communication for decision-making and data analysis purposes. Face detection, face recognition, and pose detection algorithms can be executed with acceptable performance on embedded systems and are used for home security and monitoring. However, popular machine learning frameworks, such as MediaPipe, require relatively high usage of CPU while running, even when idle with no subject in the scene. Combined with the still present false detections, this wastes CPU time, elevates the power consumption and overall system temperature, and generates unnecessary data. In this study, a low-cost low-resolution infrared thermal sensor array was used to control the execution of MediaPipe’s pose detection algorithm using single-board computers, which only runs when the thermal camera detects a possible subject in its field of view. A lightweight algorithm with several filtering layers was developed, which allowed the effective detection and isolation of a person in the thermal image. The resulting hybrid computer vision proved effective in reducing the average CPU workload, especially in environments with low activity, almost eliminating MediaPipe’s false detections, and reaching up to 30% power saving in the best-case scenario.

Next generation edge computing: A roadmap to net zero emissions

Multi-access Edge Computing (MEC, formerly, Mobile Edge Computing) offers users of Mobile Devices (MDs) such as smartphones advantages in computational task completion times by offloading to superior but geographically close computing resources. There is a need to make these services more sustainable (energy-efficient) as an initial step towards net zero emissions. We have modelled a zero-latency edge node-MD connection as a single base station linked to a devoted server with the MD transmitting and receiving data via a Wireless Local Area Network (WLAN) with, wherever possible, identifiable real-world power ratings. In this paper, we demonstrate that the total energy usage by the MD and edge node hardware can be markedly less than that of the MD alone in 3 G, 4 G and 5 G WLAN networks. The energy savings, computed as percentages of MD-alone energy usage, are independent of data file size, are greater with computationally more complex tasks, and have a higher MD CPU workload but moderate server CPU workloads. Energy savings are highly dependent on the precise configuration of the base station with server type (peak power rating, the number of CPU cores and the maximum number of jobs processed simultaneously). A 5 G base station has the highest power consumption, but this is offset by much faster WLAN speeds, which can result in energy savings in excess of 90% compared with MD computation alone. We discuss the implications of these results for reducing global electricity use and achieving carbon neutrality to contribute towards net zero emissions.

Sequential Offloading for Distributed DNN Computation in Multiuser MEC Systems

This paper studies a sequential task offloading problem for a multiuser mobile edge computing (MEC) system. While most of the existing works consider static one-shot offloading optimization with fixed wireless channel conditions and fixed computational tasks, we consider a dynamic optimization approach, which embraces wireless channel fluctuations and random deep neural network (DNN) task arrivals over an infinite horizon. Specifically, we introduce a local CPU workload queue (WD-QSI) and an MEC server workload queue (MEC-QSI) to model the dynamic workload of DNN tasks at each WD and the MEC server, respectively. The transmit power and the partitioning of the local DNN task at each WD are dynamically determined based on the instantaneous channel conditions (to capture the transmission opportunities) and the instantaneous WD-QSI and MEC-QSI (to capture the dynamic urgency of the tasks) to minimize the average latency of the DNN tasks. The joint optimization can be formulated as an ergodic Markov decision process (MDP), in which the optimality condition is characterized by a centralized Bellman equation. However, the brute force solution of the MDP is not viable due to the curse of dimensionality as well as the requirement for knowledge of the global state information. To overcome these issues, we first decompose the MDP into multiple lower dimensional sub-MDPs, each of which can be associated with a WD or the MEC server. Next, we further develop a parametric online Q-learning algorithm, so that each sub-MDP is solved locally at its associated WD or the MEC server. The proposed solution is completely decentralized in the sense that the transmit power for sequential offloading and the DNN task partitioning can be determined based on the local channel state information (CSI) and the local WD-QSI at the WD only. Additionally, no prior knowledge of the distribution of the DNN task arrivals or the channel statistics will be needed for the MEC server. The proposed solution can achieve the superb performance over various state-of-the-art baselines.

A scheduling algorithm to maximize storm throughput in heterogeneous cluster

In the most popular distributed stream processing frameworks (DSPFs), programs are modeled as a directed acyclic graph. Using this model, a DSPF can benefit from the parallelism capabilities of distributed clusters. Choosing a reasonable number of vertices for each operator and mapping the vertices to the appropriate processing resources significantly affect the overall system performance. Due to the simplicity of the current DSPF schedulers, these frameworks perform poorly on large-scale clusters. In this paper, we present a heterogeneity-aware scheduling algorithm that finds the proper number of the vertices of an application graph and maps them to the most suitable cluster node. We begin with a pre-processing step which allocates the vertices to the given cluster nodes using profiling data. Then, we gradually increase the topology input rate in order to scale up the application graph. Finally, using a CPU utilization model which predicts the CPU workload based on the input rate to vertices and the processing node’s CPU characteristics, we identify the bottlenecked vertices and allocate new instances derived from them to the least utilized processing resource. Our experimental results on Storm Micro-Benchmark show that (1) the prediction model estimate CPU utilization with 92% accuracy. (2) Compared to the default scheduler of Storm, our scheduler provides 7 to 44% throughput enhancement. (3) The proposed method can find the solution within 4% (worst case) of the optimal scheduler, which obtains the best scheduling scenario using an exhaustive search over problem design space.

Decentralized DNN Task Partitioning and Offloading Control in MEC Systems With Energy Harvesting Devices

In this paper, we study a decentralized deep neural network (DNN) task partitioning and offloading control problem for a multi-access edge computing (MEC) system with multiple wireless devices (WDs) powered by renewable energy sources. Instead of generic computational tasks, we focus on the case when the WDs and the MEC systems are computing DNN inferencing tasks. For each WD, we build a virtual local-computing CPU workload queue and an energy queue to model the DNN computation dynamics and the battery energy dynamics, respectively. We first introduce a new Lyapunov function based on the DNN tasks' cumulative execution latency and the total energy consumptions of the entire MEC system. Using a Lyapunov optimization approach, the DNN task partitioning and offloading control at the WDs are formulated as a Lyapunov drift minimization problem. To facilitate a distributed implementation, we exploit the linear structure of the cumulative latency and the independency of each WD's local-computing CPU rate, and we decompose the centralized Lyapunov drift minimization problem into multiple distributed subproblems, each of which is associated with a single WD. Next, we develop a parametric online learning algorithm such that each sub-Lyapunov drift minimization problem is solved locally at its associated WD. The proposed solution is completely decentralized in the sense that the sequential offloading and DNN task segmentation control at each WD is determined by its local battery energy queue state information (EQSI), the channel state information (CSI), and the DNN inferencing task workload queue state information (TQSI). Numerical results reveal that the proposed online learning algorithm can achieve significant performance gain over various state-of-the-art baselines.

Learning-based Phase-aware Multi-core CPU Workload Forecasting

Predicting workload behavior during workload execution is essential for dynamic resource optimization in multi-processor systems. Recent studies have proposed advanced machine learning techniques for dynamic workload prediction. Workload prediction can be cast as a time series forecasting problem. However, traditional forecasting models struggle to predict abrupt workload changes. These changes occur because workloads are known to go through phases. Prior work has investigated machine-learning-based approaches for phase detection and prediction, but such approaches have not been studied in the context of dynamic workload forecasting. In this article, we propose phase-aware CPU workload forecasting as a novel approach that applies long-term phase prediction to improve the accuracy of short-term workload forecasting. Phase-aware forecasting requires machine learning models for phase classification, phase prediction, and phase-based forecasting that have not been explored in this combination before. Furthermore, existing prediction approaches have only been studied in single-core settings. This work explores phase-aware workload forecasting with multi-threaded workloads running on multi-core systems. We propose different multi-core settings differentiated by the number of cores they access and whether they produce specialized or global outputs per core. We study various advanced machine learning models for phase classification, phase prediction, and phase-based forecasting in isolation and different combinations for each setting. We apply our approach to forecasting of multi-threaded Parsec and SPEC workloads running on an eight-core Intel Core-i9 platform. Our results show that combining GMM clustering with LSTMs for phase prediction and phase-based forecasting yields the best phase-aware forecasting results. An approach that uses specialized models per core achieves an average error of 23% with up to 22% improvement in prediction accuracy compared to a phase-unaware setup.

University of Warsaw Lagrangian Cloud Model (UWLCM) 2.0: adaptation of a mixed Eulerian–Lagrangian numerical model for heterogeneous computing clusters

Abstract. A numerical cloud model with Lagrangian particles coupled to an Eulerian flow is adapted for distributed memory systems. Eulerian and Lagrangian calculations can be done in parallel on CPUs and GPUs, respectively. The fraction of time when CPUs and GPUs work simultaneously is maximized at around 80 % for an optimal ratio of CPU and GPU workloads. The optimal ratio of workloads is different for different systems because it depends on the relation between computing performance of CPUs and GPUs. GPU workload can be adjusted by changing the number of Lagrangian particles, which is limited by device memory. Lagrangian computations scale with the number of nodes better than Eulerian computations because the former do not require collective communications. This means that the ratio of CPU and GPU computation times also depends on the number of nodes. Therefore, for a fixed number of Lagrangian particles, there is an optimal number of nodes, for which the time CPUs and GPUs work simultaneously is maximized. Scaling efficiency up to this optimal number of nodes is close to 100 %. Simulations that use both CPUs and GPUs take between 10 and 120 times less time and use between 10 to 60 times less energy than simulations run on CPUs only. Simulations with Lagrangian microphysics take up to 8 times longer to finish than simulations with Eulerian bulk microphysics, but the difference decreases as more nodes are used. The presented method of adaptation for computing clusters can be used in any numerical model with Lagrangian particles coupled to an Eulerian fluid flow.

Optimizing the EDP of OpenMP applications via concurrency throttling and frequency boosting

The increasing number of cores in multicore architectures has brought the need for better use of hardware resources. Consequently, different solutions have become widely adopted to optimize the execution of parallel applications: dynamic concurrency throttling (DCT) and turbo-boosting. While DCT artificially reduces the number of active threads to improve the performance and energy consumption, turbo-boosting increases the processor’s frequency above the base operating levels to provide maximum performance, considering the Thermal Design Power (TDP). However, as many parallel applications are unbalanced, combining both optimization knobs is not a straightforward task. Therefore, we propose TBFT. It is a transparent and automatic approach that concurrently exploits DCT and turbo-boosting to optimize OpenMP parallel applications. For that, TBFT implements a DCT Engine to find the ideal or near-ideal number of threads in parallel regions and a Boosting Engine to adjust the boosting operating mode w.r.t. the CPU workload. When executing twelve well-known benchmarks on three multicore systems, we show that TBFT outperforms different state-of-the-art strategies without jeopardizing the performance.

Novel Dynamic Scaling Algorithm for Energy Efficient Cloud Computing

Huge data processing applications are stored efficiently using cloud computing platform. Few technologies like edge computing, Internet of Things (IoT) model helps cloud computing framework for executing data with less energy and latencies for better infrastructure. Recently researches focused on providing excellent services to cloud computing users. Also, cloud-based services are highly developed over IT field. Energy a level varies based on the cloud setup like speed, memory, service capability and bandwidth. The user job requirements are varied based its nature. The process of identifying efficient energy resources for the user job is main aim of this research work. Initially IoT, Edge devices capture the job and process to help cloud infrastructure. Data transferring process is a Non-deterministic Polynomial (NP) hard problem which can be easily supported by technologies like IoT and Edge computing. Main issue is, tremendous increase of data over the cloud causes difficult to manage consumption of energy. In this research work server clouds with edge devices and centralized cloud servers are combined to work for providing efficient energy consumption. Here, in this paper we implement novel dynamic speed scaling (NDS) algorithm. The CPU workload is first computed using NDS algorithm for incoming application. The less energy computation is achieved using energy internet of things (EIoT). Fine methodology called speed processor scaling, helps to consume less energy and less computation price. High energy consumption of processor is due to high computation speed. likewise, less energy is consumed during less processor computation speed. This technique is building in NDS algorithm and computed in Edge devices for less energy consumption. The result evaluation proves that proposed technique consumes only 10 s for data computation when compared to other existing techniques.

Many-Objective Quantum-Inspired Particle Swarm Optimization Algorithm for Placement of Virtual Machines in Smart Computing Cloud.

Particle swarm optimization algorithm (PSO) is an effective metaheuristic that can determine Pareto-optimal solutions. We propose an extended PSO by introducing quantum gates in order to ensure the diversity of particle populations that are looking for efficient alternatives. The quality of solutions was verified in the issue of assignment of resources in the computing cloud to improve the live migration of virtual machines. We consider the multi-criteria optimization problem of deep learning-based models embedded into virtual machines. Computing clouds with deep learning agents can support several areas of education, smart city or economy. Because deep learning agents require lots of computer resources, seven criteria are studied such as electric power of hosts, reliability of cloud, CPU workload of the bottleneck host, communication capacity of the critical node, a free RAM capacity of the most loaded memory, a free disc memory capacity of the most busy storage, and overall computer costs. Quantum gates modify an accepted position for the current location of a particle. To verify the above concept, various simulations have been carried out on the laboratory cloud based on the OpenStack platform. Numerical experiments have confirmed that multi-objective quantum-inspired particle swarm optimization algorithm provides better solutions than the other metaheuristics.

Detection of and Countermeasure Against Thermal Covert Channel in Many-Core Systems

The thermal covert channels (TCCs) in many-core systems can cause detrimental data breaches. In this article, we present a three-step scheme to detect and fight against such TCC attacks. Specifically, in the detection step, each core calculates the spectrum of its own CPU workload traces that are collected over a few fixed time intervals, and then it applies a frequency scanning method to detect if there exists any TCC attack. In the next positioning step, the logical cores running the transmitter threads are located. In the last step, the physical CPU cores suspiciously engaging in a TCC attack have to undertake dynamic voltage frequency scaling (DVFS) such that any possible TCC trace will be essentially wiped out. Our experiments have confirmed that on average 97&#x0025; of the TCC attacks can be detected, and with the proposed defense, the packet error rate (PER) of a TCC attack can soar to more than 70&#x0025;, literally shutting down the attack in practical terms. The performance penalty caused by the inclusion of the proposed DVFS countermeasures is found to be only 3&#x0025; for an <inline-formula> <tex-math notation="LaTeX">$8\times 8$ </tex-math></inline-formula> many-core system.

Managing overloaded hosts for energy-efficiency in cloud data centers

Traditional data centers are shifted toward the cloud computing paradigm. These data centers support the increasing demand for computational and data storage that consumes a massive amount of energy at a huge cost to the cloud service provider and the environment. Considerable energy is wasted to constantly operate idle virtual machines (VMs) on hosts during periods of low load. Dynamic consolidation of VMs from overloaded or underloaded hosts is an effective strategy for improving energy consumption and resource utilization in cloud data centers. The dynamic consolidation of VM from an overloaded host directly influences the service level agreements (SLAs), utilization of resources, and quality of service (QoS) delivered by the system. We proposed an algorithm, namely, GradCent, based on the Stochastic Gradient Descent technique. This algorithm is used to develop an upper CPU utilization threshold for detecting overloaded hosts by using a real CPU workload. Moreover, we proposed a dynamic VM selection algorithm called Minimum Size Utilization (MSU) for selecting the VMs from an overloaded host for VM consolidation. GradCent and MSU maintain the trade-off between energy consumption minimization and QoS maximization under specified SLA goal. We used the CloudSim simulations with real-world workload traces from more than a thousand PlanetLab VMs. The proposed algorithms minimized energy consumption and SLA violation by 23% and 27.5% on average, compared with baseline schemes, respectively.

BHyPreC: A Novel Bi-LSTM Based Hybrid Recurrent Neural Network Model to Predict the CPU Workload of Cloud Virtual Machine

With the advancement of cloud computing technologies, there is an ever-increasing demand for the maximum utilization of cloud resources. It increases the computing power consumption of the cloud’s systems. Consolidation of cloud’s Virtual Machines (VMs) provides a pragmatic approach to reduce the energy consumption of cloud Data Centers (DC). Effective VM consolidation and VM migration without breaching Service Level Agreement (SLA) can be attained by taking proactive decisions based on cloud’s future workload prediction. Effective task scheduling, another major issue of cloud computing also relies on accurate forecasting of resource usage. Cloud workload traces exhibit both periodic and non-periodic patterns with the sudden peak of load. As a result, it is very challenging for the prediction models to precisely forecast future workload. This prompted us to propose a hybrid Recurrent Neural Network (RNN) based prediction model named BHyPreC. BHyPreC architecture includes Bidirectional Long Short-Term Memory (Bi-LSTM) on top of the stacked Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU). Here, BHyPreC is used to predict future CPU usage workload of cloud’s VM. Our proposed model enhances the non-linear data analysis capability of Bi-LSTM, LSTM, and GRU models separately and demonstrates better accuracy compared to other statistical models. The effect of variation of historical window size and training-testing data size on these models is observed. The experimental result shows that our model gives higher accuracy and performs better in comparison to Autoregressive Integrated Moving Average (ARIMA), LSTM, GRU, and Bi-LSTM model for both short-term ahead and long-term ahead prediction.

Detection of time series patterns and periodicity of cloud computing workloads

Workload pattern detection can be used as part of a proactive decision-making approach to optimize resource provisioning strategies and anticipate any performance problem in cloud computing environments. Existing workload pattern detection approaches suffer from two essential limitations which restrict their effectiveness in such environments. Specifically, they require massive human intervention, and are specific to particular types of workload data. To overcome these limitations, we propose in this paper a generic workload pattern and periodicity detection technique that employs the prefix transposition approach from the molecular biology domain to detect workload periodicity in node-specific and aggregated cloud environments. The strengths of the proposed technique compared to the state of the art lie in its ability to be applied to any type of workload, and also to detect patterns of varying lengths, amplitudes, and shapes. Experiments conducted on cloud server nodes and aggregated CPU and throughput workload datasets collected from Information Technology (IT) and Telecom domains reveal that our solution improves the accuracy of the detections, especially in harsh environments, where the lengths, shapes, and amplitudes of patterns vary, as compared to the autocorrelation technique. Other findings show that the proposed approach is also highly efficient at detecting multiple short-term and long-term periodic patterns on any type of time series-based cloud computing workloads of different time granularity.

Improving Processing Speed of Real Time Stereo Matching u sing Heterogenous CPU GPU Model

This paper presents an improvement of the processing speed of the stereo matching problem. The time required for stereo matching represents a problem for many real time applications such as robot navigation , self-driving vehicles and object tracking. In this work, a real-time stereo matching system is proposed that utilizes the parallelism of Graphics Processing Unit (GPU). An area based stereo matching system is used to generate the disparity map. Four different sequential and parallel computational models are used to analyze the time consumed by the stereo matching. The models are: 1) Sequential CPU, 2) Parallel multi-core CPU, 3) Parallel GPU and 4) Parallel heterogenous CPU/GPU. The dense disparity image is calculated, and the time is highly reduced using the heterogenous CPU/GPU model, while maintaining the same accuracy of other models. A static partitioning of CPU and GPU workload is properly designed based on time analysis. Different cost functions are used to measure the correspondence and to generate the disparity map. A sliding window is used to calculate the cost functions efficiently. A speed of more than 100 frames per second(f/s) is achieved using parallel heterogenous CPU/GPU for 640 x 480 image resolution and a disparity range equals 50.

The $$CiS^2$$: a new metric for performance and energy trade-off in consolidated servers

The increased use of cloud services has turned the virtualization in the main technology that supports cloud datacentres. To reduce the increment of power consumption caused in datacenters due to the addition of physical servers, system administrators are using virtual machine consolidation (VMC) techniques which tries to allocate the adequate number of virtual machines per physical server. Therefore, VMC increases server resources utilization and as a consequence its performance degradation and the energy consumption too. Then, a trade-off between the performance and the energy consumption exists when consolidating virtual machines. This trade-off is difficult to quantify and also to determine the servers efficiency taking into account a specific number of allocated virtual machines. Because of this, it is crucial for system administrators having a simple metric that assists the VMC making-decision process. In this paper, we propose the $$CiS^2$$ index, a metric to quantify this performance-energy trade-off. Also, this index can help system administrators to decide about the servers’ efficiency through benchmarking and to select the most efficient server through a proposed algorithm. Besides, we propose a simple graphical representation of the index to distinguish graphically the efficient and non-efficient server consolidations. We validate the index in a theoretical manner and performing real experiments in different physical servers under CPU workload saturation. Obtained results show that the proposed index reflects the performance-energy trade-off behaviour and it helps systems’ administrators when consolidating virtual machines.

Modeling Real-World Load Patterns for Benchmarking in Clouds and Clusters

Cloud computing has currently permeated all walks of life. It has proven extremely useful for organizations and individual users to save costs by leasing compute resources that they need. This has led to an exponential growth in cloud computing based research and development. A substantial number of frameworks, approaches and techniques are being proposed to enhance various aspects of clouds, and add new features. One of the constant concerns in this scenario is creating a testbed that successfully reflects a real-world cloud datacenter. It is vital to simulate realistic, repeatable, standardized CPU and memory workloads to compare and evaluate the impact of the different approaches in a cloud environment. This paper introduces Cloudy, which is an open-source workload generator that can be used within cloud instances, Virtual Machines (VM), containers, or local hosts. Cloudy utilizes resource usage traces of machines from Google and Alibaba clusters to simulate up to 16000 different, real-world CPU and memory load patterns. The tool also provides a variety of machine metrics for each run, that can be used to evaluate and compare the performance of the VM, container or host. Additionally, it includes a web-based visualization component that offers a number of real-time statistics, as well as overall statistics of the workload such as seasonal trends, and autocorrelation. These statistics can be used to further analyze the real-world traces, and enhance the understanding of workloads in the cloud.