Storage Savings Research Articles

Efficient Convolution-Based Representation of Smooth Reference Trajectories for Urban Air Mobility

Accurate trajectory-following control using dynamic inversion controllers requires suitable parameterizations of reference trajectories, such as energy-minimal flight paths generated with optimal control methods. These trajectory representations must enable efficient onboard storage and real-time evaluation while ensuring sufficient accuracy and smoothness to provide the required feedforward commands. We explore trajectory parameterizations based on the convolution of a piecewise low-order polynomial with an infinitely smooth kernel. These combine efficient storage with unlimited differentiability and allow for simple and efficient evaluation using numerical quadrature. To construct such parameterizations from given trajectories, we formulate quadratic programs that account for additional, mission-specific constraints. A numerical example demonstrates the parameterization of an approach and vertical landing trajectory based on optimal control results; we show that the convolution-based approach yields competitive results compared to a piecewise polynomial parameterization, with significant storage savings.

Mixed Palletizing for Smart Warehouse Environments: Sustainability Review of Existing Methods

Mixed palletizing constitutes one of the problems in the logistics domain aroused from the need for fast product movement to satisfy the continuously increasing number of customers with the demand for highly personalized goods. In a demanding environment, such as warehouses, where break bulk and the consolidation of loads take up most of the working time, the automation of mixed palletizing can lead to increased efficiency and speed while keeping errors to a minimum. Space utilization of pallets enables savings in storage and transportation costs, boosting the overall sustainable role of the Supply Chain sector. This paper presents the proposed approaches to mixed palletizing stemming from different fields, with a focus on recent developments in the application of Industry 4.0 technologies. Our research highlights quite a few areas that require attention from researchers.

Data cube-based storage optimization for resource-constrained edge computing

In the evolving landscape of the digital era, edge computing emerges as an essential paradigm, especially critical for low-latency, real-time applications and Internet of Things (IoT) environments. Despite its advantages, edge computing faces severe limitations in storage capabilities and is fraught with reliability issues due to its resource-constrained nature and exposure to challenging conditions. To address these challenges, this work presents a tailored storage mechanism for edge computing, focusing on space efficiency and data reliability. Our method comprises three key steps: relation factorization, column clustering, and erasure encoding with compression. We successfully reduce the required storage space by deconstructing complex database tables and optimizing data organization within these sub-tables. We further add a layer of reliability through erasure encoding. Comprehensive experiments on TPC-H datasets substantiate our approach, demonstrating storage savings of up to 38.35% and time efficiency improvements by 3.96x in certain cases. Furthermore, our clustering technique shows a potential for additional storage reduction up to 40.41%.

Effect of Reduced Oil Flow Rate on the Performance of a Load on Pad Journal Bearing: Flooded Versus Evacuated Conditions

Abstract Means to decrease the energy consumption of tilting pad journal bearings (TPJBs) without affecting their performance and structural integrity are mandatory in a cost efficient operation. Reducing the lubricant flow supplied to a bearing is a distinct method to diminish drag power losses along with savings in oil storage and pump equipment. However, a too low flow remains questionable in industrial practice. Starved flow conditions produce hot pad surfaces that could lead to Babbitt failure; and under certain loads, generate subsynchronous vibrations (SSV). This paper aims to resolve some of the issues above via measurements of the load performance conducted with a four-pad TPJB configured as load-on-pad (LOP) and having its ends sealed or open to make flooded and evacuated conditions. Prior measurements were conducted with the same bearing under load-between-pads (LBP); see Refs. [1] and [2]. The nominal supplied flow (Q) of ISO VG 46 oil at 60 °C is proportional to shaft speed (max. 12 krpm: 62.8 m/s surface speed). In the tests, the flow Qs ranges from 1.5 Q to just ¼ Q, and the applied units load reaches 2.07 MPa. Compared to the flooded bearing, the evacuated bearing produces a slightly larger eccentricity across the range of flow rates. For a unit load = 2.07 MPa and shaft speed of 6 or 12 krpm, the highest pad subsurface temperature reaches ∼130 °C for Qs below 50% nominal. For both bearings, flooded or evacuated, drag power losses decrease to ∼30% as the oil flow drops from 100% to 50% Q. As the oil flow increases to 1.5 Q, the drag power increases ∼10% for both bearing types at 6 krpm and for the flooded one at 12 krpm, while the evacuated bearing shows a reduction of ∼7%. The drag power grows as the static load increases; the evacuated bearing producing up to ∼ 40% lesser power loss than the flooded bearing. Both bearings produce similar size direct stiffnesses though largely orthotropic, Kyy ≫ Kxx. Direct damping coefficients Cxx ∼ Cyy increase with shaft speed and unit load but dramatically decrease as Qs drops, in particular for the evacuated bearing. The current measurements and those in Refs. [1] and [2] demonstrate that LOP and LBP bearings can safely operate with 50% of nominal flow thus saving drag power, and without too large pad, temperature rises. Alas, too low Qs produces a significant reduction in the damping coefficient orthogonal to the applied load direction. The effect is most evident in the LOP evacuated bearing, which is most prone to show SSV Hash.

Coarsening effects on k-partite network classification

The growing data size poses challenges for storage and computational processing time in semi-supervised models, making their practical application difficult; researchers have explored the use of reduced network versions as a potential solution. Real-world networks contain diverse types of vertices and edges, leading to using k-partite network representation. However, the existing methods primarily reduce uni-partite networks with a single type of vertex and edge. We develop a new coarsening method applicable to the k-partite networks that maintain classification performance. The empirical analysis of hundreds of thousands of synthetically generated networks demonstrates the promise of coarsening techniques in solving large networks’ storage and processing problems. The findings indicate that the proposed coarsening algorithm achieved significant improvements in storage efficiency and classification runtime, even with modest reductions in the number of vertices, leading to over one-third savings in storage and twice faster classifications; furthermore, the classification performance metrics exhibited low variation on average.

Practical Design Considerations for Wide Locally Recoverable Codes (LRCs)

Most of the data in large-scale storage clusters is erasure coded. At exascale, optimizing erasure codes for low storage overhead, efficient reconstruction, and easy deployment is of critical importance. Locally recoverable codes (LRCs) have deservedly gained central importance in this field, because they can balance many of these requirements. In our work, we study wide LRCs; LRCs with large number of blocks per stripe and low storage overhead. These codes are a natural next step for practitioners to unlock higher storage savings, but they come with their own challenges. Of particular interest is their reliability , since wider stripes are prone to more simultaneous failures. We conduct a practically minded analysis of several popular and novel LRCs. We find that wide LRC reliability is a subtle phenomenon that is sensitive to several design choices, some of which are overlooked by theoreticians, and others by practitioners. Based on these insights, we construct novel LRCs called Uniform Cauchy LRCs , which show excellent performance in simulations and a 33% improvement in reliability on unavailability events observed by a wide LRC deployed in a Google storage cluster. We also show that these codes are easy to deploy in a manner that improves their robustness to common maintenance events. Along the way, we also give a remarkably simple and novel construction of distance-optimal LRCs (other constructions are also known), which may be of interest to theory-minded readers.

Enabling Efficient Deduplication and Secure Decentralized Public Auditing for Cloud Storage: A Redactable Blockchain Approach

Public auditing and data deduplication are integral considerations in providing efficient and secure cloud storage services. Nevertheless, the traditional data deduplication models that support public auditing can endure the enormous waste of storage and computation resources induced through data redundancy and repeated audit work by multiple tenants on trusted third-party auditor (TPA). In this work, we introduce blockchain-based secure decentralized public auditing in a decentralized cloud storage with an efficient deduplication model. We employ blockchain to take on the task of centralized TPA, which also mitigates the implications of malicious blockchain miners by using the concept of a decentralized autonomous organization (DAO). Specifically, we employ the idea of redactability for blockchain to handle often neglected security issues that would adversely affect the integrity of stored auditing records on blockchain in decentralized auditing models. However, the proposed model also employs an efficient deduplication scheme to attain adequate storage savings while preserving the users from data loss due to duplicate faking attacks. Moreover, the detailed concrete security analysis demonstrates the computational infeasibility of the proposed model against proof-of-ownership, duplicate faking attack (DFA), collusion attack, storage free-riding attack, data privacy, and forgery attack with high efficiency. Finally, the comprehensive performance analysis shows the scalability and feasibility of the proposed model.

Modewise Johnson–Lindenstrauss embeddings for nuclear many-body theory

In this work, we explore modewise Johnson-Lindenstrauss embeddings (JLEs) as a tool to reduce the computational cost and memory requirements of nuclear many-body methods. JLEs are randomized projections of high-dimensional data tensors onto low-dimensional subspaces that preserve key structural features. Such embeddings allow for the oblivious and incremental compression of large tensors, e.g., the nuclear Hamiltonian, into significantly smaller random sketches that still allow for the accurate calculation of ground-state energies and other observables. Their oblivious character makes it possible to compress a tensor without knowing in advance exactly what observables one might want to approximate at a later time. This opens the door for the use of tensors that are much too large to store in memory, e.g., complete two-plus three-nucleon Hamiltonians in large, symmetry-unrestricted bases. Such compressed Hamiltonians can be stored and used later on with relative ease. As a first step, we analyze the JLE's impact on the second-order Many-Body Perturbation Theory (MBPT) corrections for nuclear ground-state observables. Numerical experiments for a wide range of closed-shell nuclei, model spaces and state-of-the-art nuclear interactions demonstrate the validity and potential of the proposed approach: We can compress nuclear Hamiltonians hundred- to thousand-fold while only incurring mean relative errors of 1\% or less in ground-state observables. Importantly, we show that JLEs capture the relevant physical information contained in the highly structured Hamiltonian tensor despite their random characteristics. In addition to the significant storage savings, the achieved compressions imply multiple order-of magnitude reductions in computational effort when the compressed Hamiltonians are used in higher-order MBPT or nonperturbative many-body methods.

Design And Fabrication of Experimental Manufacturing of Pellets from Waste Material

Many developing countries produce huge amount of agro waste. The disposal of this waste is a major problem. The residues like wheat and rice husk, sawdust from carpentry shops, groundnut shells, and dry leaves from plants are available in a large quantity. Apart from the problems of storage, transportation and easy handling of this waste, burning of biomass in grates poses a greater problem of widespread air pollution. Biomass pellets are a well-liked sort of biomass fuel, generally made up of wood wastes, agricultural biomass, commercial grasses and forestry residues. In addition to savings in transportation and storage, pelletization of biomass facilitates easy and price effective handling. Dense cubes pellets have the flow ability characteristics almost like those of cereal grains. The even geometry and structure of biomass pellet helps in automatic feeding. Densely packed biomass pellets help in good durability in transport and storage. Pellets if densely packed and the moisture content is low the combustible properties can be increased. Keywords: Solar Energy, Wind Energy, PV Cell, Renewable Energy, Hybrid Power System, Electricity.

Evaluation of a new apple in-field sorting system for fruit singulation, rotation and imaging

Automated, computer-imaging based in-field grading and sorting of apples at the time of harvest has the potential to help growers achieve significant cost savings in postharvest storage and packing. Singulation and rotation of fruit are essential steps for a computer imaging-based grading/sorting system. We have recently developed a novel compact, low-cost in-field apple sorting system, which consists of pairs of multi-stage helical conveyance drives to singulate, rotate and advance apples so that they can be inspected by a computer vision unit for quality grading. This study was aimed at evaluating the singulating and rotating performance of the sorting system for automatic inspection and grading of apples, in terms of conveying speed, fruit size, and the percentage of apple exposure area for imaging. Two apple varieties (i.e., ‘Golden Delicious’ and ‘Delicious’) were used for experimental evaluations. To determine the rotational behavior of apples, each quadrant of the apple surface was painted with different colors. Images were acquired for the apples at 15 frames per second and then automatically segmented for the estimation of exposed areas in terms of pixels. It was found that the number of images acquired for each fruit varied from 24 to 9 images when the speed of the sorting system was changed from 1 to 3 apples per second per lane. The singulating and rotating mechanism (SRM) provided continuous and relatively even rotation of the apples while conveying, which was confirmed by cumulative percentages of the four colored areas of ‘Golden Delicious’ and ‘Delicious’ apples at the three sorting speeds. At the sorting speed of 1 apple/s, the entire surface of each fruit was imaged for more than 4 times, while at 3 apples/s, the surface of the apples was imaged for at least 1.4 times. Hence, the sorting system with 3 sorting lanes is able to perform in-field quality grading of apples at a throughput of 9 or more apples/s.

Dynamic ConvNets on Tiny Devices via Nested Sparsity

This work introduces a new training and compression pipeline to build nested sparse convolutional neural networks (ConvNets), a class of dynamic ConvNets suited for inference tasks deployed on resource-constrained devices at the edge of the Internet of Things. A nested sparse ConvNet consists of a single ConvNet architecture, containing <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$N$ </tex-math></inline-formula> sparse subnetworks with nested weights subsets, like a Matryoshka doll, and can trade accuracy for latency at runtime, using the model sparsity as a dynamic knob. To attain high accuracy at training time, we propose a gradient masking technique that optimally routes the learning signals across the nested weight subsets. To minimize the storage footprint and efficiently process the obtained models at inference time, we introduce a new sparse matrix compression format with dedicated compute kernels that fruitfully exploit the characteristic of the nested weights subsets. Tested on image classification and object detection tasks on an off-the-shelf ARM-M7 microcontroller unit (MCU), nested sparse ConvNets outperform variable-latency solutions naively built assembling single sparse models trained as stand-alone instances, achieving 1) comparable accuracy; 2) remarkable storage savings; and 3) high performance. Moreover, when compared to state-of-the-art dynamic strategies, such as dynamic pruning and layer width scaling, nested sparse ConvNets turn out to be Pareto optimal in the accuracy versus latency space.

CAQ: Toward Context-Aware and Self-Adaptive Deep Model Computation for AIoT Applications

Artificial Intelligence of Things (AIoT) has recently accepted significant interests. Remarkably, embedded artificial intelligence (e.g., deep learning) on-device transforms IoT devices into intelligent systems that robustly and privately process data. Quantization technique is widely used to compress deep models for narrowing the resource gap between computation demands and platform supply. However, existing quantization schemes induce unsatisfaction for IoT scenarios since they are oblivious to dynamic changes of application context (e.g., battery and hierarchical memory availability) during the long-term operation. Subsequently, they will mismatch the user-desired resource efficiency and application lifetime. Also, to adapt to the dynamic context, we can neither accept the latency for model retraining with existing hand-crafted quantization nor the overhead for quantization bit width researching with prior on-demand quantization. This article presents a context-aware and self-adaptive deep model quantization (CAQ) system for IoT application scenarios. CAQ integrates a novel switchable multigate quantization framework, optimizing the quantized model accuracy and energy efficiency in diverse contexts. Based on the learned model, CAQ can switch among different gating networks in a context-aware manner and then adopt it to automatically capture the representation importance of various layers for optimal quantization bit-width selection. The experimental results show that CAQ achieves up to 50% storage savings with even 2.61% higher accuracy than the state-of-the-art baselines.

Using raft as consensus algorithm for blockchain application of roaming services for mobile network

In the past few years blockchain became the new trending technology, it is expected to spread as internet in the 80s, Blockchain technology, using a consensus algorithm, is a cryptographic record that cannot be changed by repetitive sequence hashing and failure tolerance. Decentralization, persistency, anonymity, and audibility is the key characteristics of the blockchain. These characteristics encourage many industries to be interested in blockchain adoption in their IT systems. The consensus is one of the critical problems in the blockchain; these consensus algorithms play a critical role in maintaining the safety and efficiency of the blockchain. Unless transactions are pending, the Raft Consensus will not mint blocks. The raft can lead to significant storage savings, particularly if the transaction load is low because zero transactions are not minted in empty blocks. The performance of a blockchain that implements raft as a consensus algorithm has been evaluated through the simulation of roaming services for mobile network operators (MNOs) application. Performance evaluation to determine how quickly the network accepts blockchain transactions are reported. The result indicated that the proposed mechanism increase the throughput of transaction significantly.

A Structural Subspace Clustering Approach for Hyperspectral Band Selection

Band selection, which removes irrelevant bands from hyperspectral images (HSIs) and keeps essential spectral information contained in a relatively few bands, allows huge savings in data storage, computation time, and imaging hardware. In this article, we propose a novel structural subspace clustering (STSC) method for hyperspectral band selection, which leverages the self-representation property of data and structural prior information to learn the cluster structure of bands. Particularly, we propose a general clustering model where the coarse coefficients matrix derived from a self-representation model is decomposed as a combination of a desirable coefficients matrix and a sparse matrix. This strategy adaptively adjusts the coarse coefficients matrix to learn the intrinsic data structure in low-dimensional subspaces. To guide this learning process, we introduce a structural regularization approach which makes use of the prior information about local and global properties of spectral bands. Moreover, we incorporate also prior knowledge about the dictionary, which demonstrates to yield a better clustering performance. We develop an adaptive method to estimate the number of selected bands by analyzing eigenvalue gaps of Laplacian matrix. To solve the resulting model, an efficient algorithm based on alternating direction method of multipliers (ADMMs) is developed. Extensive experiments on benchmark HSIs show that our method outperforms the state-of-the-art band selection methods.

Hardware Adaptive High‐Order Interpolation for Real‐Time Graphics

AbstractInterpolation is a core operation that has widespread use in computer graphics. Though higher‐order interpolation provides better quality, linear interpolation is often preferred due to its simplicity, performance, and hardware support.We present a unified refactoring of quadratic and cubic interpolations as standard linear interpolation plus linear interpolations of higher‐order terms and show how they can be applied to regular grids and (triangular/tetrahedral) simplexes Our formulations can provide significant reduction in computation cost, as compared to typical higher‐order interpolations and prior approaches that utilize existing hardware linear interpolation support to achieve higher‐order interpolation. In addition, our formulation allows approximating the results by dynamically skipping some higher order terms with low weights for further savings in both computation and storage. Thus, higher‐order interpolation can be performed adaptively, as needed.We also describe how relatively minor modifications to existing GPU hardware could provide hardware support for quadratic and cubic interpolations using our approach for both texture filtering operations and barycentric interpolation.We present a variety of examples using triangular, rectangular, tetrahedral, and cuboidal interpolations, showing the effectiveness of our higher‐order interpolations in different applications.

Three-Dimensional Central Moment Lattice Boltzmann Method on a Cuboid Lattice for Anisotropic and Inhomogeneous Flows

Lattice Boltzmann (LB) methods are usually developed on cubic lattices that discretize the configuration space using uniform grids. For efficient computations of anisotropic and inhomogeneous flows, it would be beneficial to develop LB algorithms involving the collision-and-stream steps based on orthorhombic cuboid lattices. We present a new 3D central moment LB scheme based on a cuboid D3Q27 lattice. This scheme involves two free parameters representing the ratios of the characteristic particle speeds along the two directions with respect to those in the remaining direction, and these parameters are referred to as the grid aspect ratios. Unlike the existing LB schemes for cuboid lattices, which are based on orthogonalized raw moments, we construct the collision step based on the relaxation of central moments and avoid the orthogonalization of moment basis, which leads to a more robust formulation. Moreover, prior cuboid LB algorithms prescribe the mappings between the distribution functions and raw moments before and after collision by using a moment basis designed to separate the trace of the second order moments (related to bulk viscosity) from its other components (related to shear viscosity), which lead to cumbersome relations for the transformations. By contrast, in our approach, the bulk and shear viscosity effects associated with the viscous stress tensor are naturally segregated only within the collision step and not for such mappings, while the grid aspect ratios are introduced via simpler pre- and post-collision diagonal scaling matrices in the above mappings. These lead to a compact approach, which can be interpreted based on special matrices. It also results in a modular 3D LB scheme on the cuboid lattice, which allows the existing cubic lattice implementations to be readily extended to those based on the more general cuboid lattices. To maintain the isotropy of the viscous stress tensor of the 3D Navier–Stokes equations using the cuboid lattice, corrections for eliminating the truncation errors resulting from the grid anisotropy as well as those from the aliasing effects are derived using a Chapman–Enskog analysis. Such local corrections, which involve the diagonal components of the velocity gradient tensor and are parameterized by two grid aspect ratios, augment the second order moment equilibria in the collision step. We present a numerical study validating the accuracy of our approach for various benchmark problems at different grid aspect ratios. In addition, we show that our 3D cuboid central moment LB method is numerically more robust than its corresponding raw moment formulation. Finally, we demonstrate the effectiveness of the 3D cuboid central moment LB scheme for the simulations of anisotropic and inhomogeneous flows and show significant savings in memory storage and computational cost when used in lieu of that based on the cubic lattice.

Where did my 256 GB go? A Measurement Analysis of Storage Consumption on Smart Mobile Devices

This work presents the first-ever detailed and large-scale measurement analysis of storage consumption behavior of applications (apps) on smart mobile devices. We start by carrying out a five-year longitudinal static analysis of millions of Android apps to study the increase in their sizes over time and identify various sources of app storage consumption. Our study reveals that mobile apps have evolved as large monolithic packages that are packed with features to monetize/engage users and optimized for performance at the cost of redundant storage consumption. We also carry out a mobile storage usage study with 140 Android participants. We built and deployed a lightweight context-aware storage tracing tool, called cosmos, on each participant's device. Leveraging the traces from our user study, we show that only a small fraction of apps/features are actively used and usage is correlated to user context. Our findings suggest a high degree of app feature bloat and unused functionality, which leads to inefficient use of storage. Furthermore, we found that apps are not constrained by storage quota limits, and developers freely abuse persistent storage by frequently caching data, creating debug logs, user analytics, and downloading advertisements as needed. Finally, drawing upon our findings, we discuss the need for efficient mobile storage management, and propose an elastic storage design to reclaim storage space when unused. We further identify research challenges and quantify expected storage savings from such a design. We believe our findings will be valuable to the storage research community as well as mobile app developers.

Where did my 256 GB go? A Measurement Analysis of Storage Consumption on Smart Mobile Devices

This work presents the first-ever detailed and large-scale measurement analysis of storage consumption behavior of applications (apps) on smart mobile devices. We start by carrying out a five-year longitudinal static analysis of millions of Android apps to study the increase in their sizes over time and identify various sources of app storage consumption. Our study reveals that mobile apps have evolved as large monolithic packages that are packed with features to monetize/engage users and optimized for performance at the cost of redundant storage consumption. We also carry out a mobile storage usage study with 140 Android participants. We built and deployed a lightweight context-aware storage tracing tool, called cosmos, on each participant's device. Leveraging the traces from our user study, we show that only a small fraction of apps/features are actively used and usage is correlated to user context. Our findings suggest a high degree of app feature bloat and unused functionality, which leads to inefficient use of storage. Furthermore, we found that apps are not constrained by storage quota limits, and developers freely abuse persistent storage by frequently caching data, creating debug logs, user analytics, and downloading advertisements as needed. Finally, drawing upon our findings, we discuss the need for efficient mobile storage management, and propose an elastic storage design to reclaim storage space when unused. We further identify research challenges and quantify expected storage savings from such a design. We believe our findings will be valuable to the storage research community as well as mobile app developers.

Lorentz Boosts and Wigner Rotations: Self-Adjoint Complexified Quaternions

In this paper, Lorentz boosts and Wigner rotations are considered from a (complexified) quaternionic point of view. It is demonstrated that, for a suitably defined self-adjoint complex quaternionic 4-velocity, pure Lorentz boosts can be phrased in terms of the quaternion square root of the relative 4-velocity connecting the two inertial frames. Straightforward computations then lead to quite explicit and relatively simple algebraic formulae for the composition of 4-velocities and the Wigner angle. The Wigner rotation is subsequently related to the generic non-associativity of the composition of three 4-velocities, and a necessary and sufficient condition is developed for the associativity to hold. Finally, the authors relate the composition of 4-velocities to a specific implementation of the Baker–Campbell–Hausdorff theorem. As compared to ordinary 4×4 Lorentz transformations, the use of self-adjoint complexified quaternions leads, from a computational view, to storage savings and more rapid computations, and from a pedagogical view to to relatively simple and explicit formulae.

Building enclave-native storage engines for practical encrypted databases

Data confidentiality is one of the biggest concerns that hinders enterprise customers from moving their workloads to the cloud. Thanks to the trusted execution environment (TEE), it is now feasible to build encrypted databases in the enclave that can process customers' data while keeping it confidential to the cloud. Though some enclave-based encrypted databases emerge recently, there remains a large unexplored area in between about how confidentiality can be achieved in different ways and what influences are implied by them. In this paper, we first provide a broad exploration of possible design choices in building encrypted database storage engines, rendering trade-offs in security, performance and functionality. We observe that choices on different dimensions can be independent and their combination determines the overall trade-off of the entire storage. We then propose Enclage , an encrypted storage engine that makes practical trade-offs. It adopts many enclave-native designs, such as page-level encryption, reduced enclave interaction, and hierarchical memory buffer, which offer high-level security guarantee and high performance at the same time. To make better use of the limited enclave memory, we derive the optimal page size in enclave and adopt delta decryption to access large data pages with low cost. Our experiments show that Enclage outperforms the baseline, a common storage design in many encrypted databases, by over 13x in throughput and about 5x in storage savings.