Real-world Workloads Research Articles

Design and Implementation of Deduplication on F2FS

Data deduplication technology has gained popularity in modern file systems due to its ability to eliminate redundant writes and improve storage space efficiency. In recent years, the flash-friendly file system (F2FS) has been widely adopted in flash memory-based storage devices, including smartphones, fast-speed servers, and Internet of Things. In this article, we propose F2DFS (deduplication-based F2FS), which introduces three main design contributions. First, F2DFS integrates inline and offline hybrid deduplication. Inline deduplication eliminates redundant writes and enhances flash device endurance, while offline deduplication mitigates the negative I/O performance impact and saves more storage space. Second, F2DFS follows the file system coupling design principle, effectively leveraging the potentials and benefits of both deduplication and native F2FS. Also, with the aid of this principle, F2DFS achieves high-performance and space-efficient incremental deduplication. Third, F2DFS adopts virtual indexing to mitigate deduplication-induced many-to-one mapping updates during the segment cleaning. We conducted comprehensive experimental comparisons between F2DFS, native F2FS, and other state-of-the-art deduplication schemes, using both synthetic and real-world workloads. For inline deduplication, F2DFS outperforms SmartDedup, Dmdedup, and ZFS, in terms of both I/O bandwidth performance and deduplication rates. And for offline deduplication, compared to SmartDedup, XFS, and BtrFS, F2DFS shows higher execution efficiency, lower resource usage, and greater storage space savings. Moreover, F2DFS demonstrates more efficient segment cleanings than native F2FS.

Extremely-Compressed SSDs with I/O Behavior Prediction

As the data volume continues to grow exponentially, there is an increasing demand for large storage system capacity. Data compression techniques effectively reduce the volume of written data, enhancing space efficiency. As a result, many modern SSDs have already incorporated data compression capabilities. However, data compression introduces additional processing overhead in critical I/O paths, potentially affecting system performance. Currently, most compression solutions in flash-based storage systems employ fixed compression algorithms for all incoming data without leveraging differences among various data access patterns. This leads to sub-optimal compression efficiency. This article proposes a data-type-aware Flash Translation Layer (DAFTL) scheme to maximize space efficiency without compromising system performance. First, we propose an I/O behavior prediction method to forecast future access on specific data. Then, DAFTL matches data types with distinct I/O behaviors to compression algorithms of varying intensities, achieving an optimal balance between performance and space efficiency. Specifically, it employs higher-intensity compression algorithms for less frequently accessed data to maximize space efficiency. For frequently accessed data, it utilizes lower-intensity but faster compression algorithms to maintain system performance. Finally, an improved compact compression method is proposed to effectively eliminate page fragmentation and further enhance space efficiency. Extensive evaluations using a variety of real-world workloads, as well as the workloads with real data we collected on our platforms, demonstrate that DAFTL achieves more data reductions than other approaches. When compared to the state-of-the-art compression schemes, DAFTL reduces the total number of pages written to the SSD by an average of 8%, 21.3%, and 25.6% for data with high, medium, and low compressibility, respectively. In the case of workloads with real data, DAFTL achieves an average reduction of 10.4% in the total number of pages written to SSD. Furthermore, DAFTL exhibits comparable or even improved read and write performance compared to other solutions.

KGFabric: A Scalable Knowledge Graph Warehouse for Enterprise Data Interconnection

Based on the diversified application scenarios at Ant Group, we built the Ant Knowledge Graph Platform (AKGP). It has constructed numerous domain-specific knowledge graphs related to merchants, companies, accounts, products, and more. AKGP manages trillions of structured knowledge graphs, serving search, recommendation, risk control and other businesses. However, as the demand increasing for various workloads such as graph pattern matching, graph representation learning, and cross-domain knowledge reuse, the existing warehouse systems based on relational DBMS or graph databases are unable to meet the requirements. To address these issues, we propose KGFabric, an industrial-scale knowledge graph management system built on the distributed file system (DFS). KGFabric offers a nearline knowledge storage engine that utilizes a Semantic-enhanced Programmable Graph (SPG) model, which is compatible with the Labeled Property Graph (LPG) model. The data is persistently stored in DFS, such as HDFS, which leverages the POSIX file system API, making it suitable for deployment in multi-cloud environment at low cost. KGFabric provides a native graph-based and hybrid storage format that can serve as a shared backend for parallel graph computing systems, significantly accelerating the analysis of multi-workload. Additionally, KGFabric includes a graph fabric framework that minimizes data duplication and guarantees data security. KGFabric is able to manage Peta-scale data and has supported graph fabric and analysis with over 100 billion relations at Ant Group. We conduct experiments on various datasets to evaluate the performance of KGFabric. Compared with popular relational DBMS and graph databases, the storage space for semantic relations is reduced by over 90%. The performance of graph fabric improves by 21× in real-world workloads. In multi-hop semantic graph analysis, KGFabric enhances performance by 100×.

Optimasi dan Modifikasi Debian untuk Meningkatkan Kinerja Sistem Operasi Real-Time

In today's digital era, operating systems play a crucial role in various technological devices. This paper discusses the optimization and modification of Debian, one of the most stable Linux distributions, to enhance its performance as a real-time operating system (RTOS). The primary objectives of this research are to improve the responsiveness, reliability, and stability of Debian in handling real-time tasks. The study involves modifying the kernel, adjusting scheduling settings, optimizing memory management, and integrating specific real-time software. The results of the tests show that the modified Debian provides lower and more stable task execution latency compared to the standard Debian. Tests were conducted using the cyclictest software to measure system latency and Htop to monitor CPU and memory performance. Additionally, the use of the VLC application as a real-world workload demonstrated that the optimized operating system could handle task priorities more efficiently, allocate resources as needed for real-time demands, and maintain stability under heavy workloads. This research significantly contributes to the development of Debian-based RTOS, which can be applied in fields such as industrial automation, robotics, and IoT devices. The optimizations ensure that the operating system can meet the high-performance and real-time reliability demands of these critical applications.

HRMF-DRP: A next-generation solution for overcoming provisioning challenges in cloud environments

The cloud computing infrastructure is a distributed environment and the existing research works have some provisioning problems such as suboptimal resource utilization and high execution time. The Heterogeneity Resource Management Framework for Dynamic Resource Provisioning (HRMF-DRP) is proposed for focusing on task scheduling and workload management. This framework incorporates advanced algorithms for dataset preprocessing, task clustering, workload prediction, and dynamic resource provisioning. For data preprocessing, the real-world workload traces were captured from the Planet Lab dataset that are taken as input for the preprocessing stage. The data preprocessing is responsible for ensuring data quality and reliability by using different models like missing data handling, outlier detection and removal as well as standardization and normalization. In this paper, the tasks are grouped into clusters by utilizing Density-Based Spatial Clustering of Applications with Noise (DBSCAN) model and this model categorizes the data points into border points, core points and noise points based on their density. The temporal dependencies are captured for the workload prediction by using Long Short-Term Memory (LSTM) neural network model. A Gaussian Mixture Model (GMM) model is responsible for estimating the number of Virtual machines (VMs) present in the workload prediction process. The Self-Adaptive Genetic Algorithm (SAGA) is implemented for task mapping that adjusts the parameters to change workload patterns for contributing adaptability and robustness. The different experimental evaluations are conducted based on the task completion time, workload balance index, resource utilization efficiency and workload prediction accuracy. The proposed model achieved the workload prediction accuracy of 98.5%, cost of $89.6, execution time of 125ms, Task Completion Time (TCT) of 40ms, Workload Balance Index (WBI) of 0.96 and Resource Utilization Efficiency (RUE) of 0.93. The quantitative results collectively position HRMF-DRP as a practical and efficient solution, promising advancements in dynamic resource provisioning for cloud computing, particularly within the Infrastructure as a Service (IaaS) cloud model.

Energy-aware virtual machine placement based on a holistic thermal model for cloud data centers

As energy-intensive infrastructures, data centers (DCs) have become a pressing challenge for managers due to their significant energy consumption and carbon emissions. Information technology (IT) and cooling systems contribute the most to energy consumption. Energy-aware virtual machine (VM) scheduling methods have been widely demonstrated to reduce energy consumption and operating costs in DCs. However, as realistic DCs exhibit complex power and thermodynamic behaviors, existing works cannot provide efficient measures to optimize computing and cooling power consumption simultaneously. To overcome this challenge, we construct a holistic thermal model (including CPU and server inlet thermal models) to accurately represent the non-uniform, dynamic thermal environment. Subsequently, this work proposes a thermal model-based energy-aware VM placement method (TEVP) to minimize the holistic energy consumption of the DCs, considering resource and thermal constraints. We develop a novel hybrid swarm intelligence algorithm (DE-ERPSO) combining differential evolution (DE) and particle swarm optimization with an elite re-selection mechanism (ERPSO) to explore more energy-efficient VM placement schemes. Extensive experiments are conducted on an extended CloudSim to validate the performance of the proposed TEVP using real-world workload traces (PlanetLab and Azure). Results show that TEVP saves over 5.6% of the total energy consumption over the advanced baselines while maintaining low thermal violations.

OCL-MEC: An online CPU-core prediction based on load balancing framework for offloading resource management in mobile edge computing environment

Clients can increase or decrease the number of resources they use dynamically over time due to the elasticity of cloud resources. As a result, variations in resource demands and predefined VM sizes result in a lack of resource utiliation, load imbalances, and excessive power consumption. A framework of efficient resource management is proposed to address these issues, balancing the load accordingly and anticipating the resource utilization of the servers. By optimizing resource utilization and minimizing the number of active servers, this technique facilitates power savings. Under/overloaded servers reduce energy consumption, execution delay, and performance degradation through a resource prediction system that is deployed at the CPU. Moreover, OCL-MEC load-balancing and resource allocation algorithms are proposed to reduce data center network traffic and power consumption. Experiments on real-world workload datasets, namely Bitsbrain VM traces, are conducted to evaluate the proposed framework. Different performance metrics demonstrate the superiority of the proposed framework over state-of-the-art approaches. Power savings of up to 98 % can be achieved by the OCL-MEC framework using a decision tree load balancing model based on HMM prediction systems.

Learning complex predicates for cardinality estimation using recursive neural networks

Cardinality estimation is one of the most vital components in the query optimizer, which has been extensively studied recently. On one hand, traditional cardinality estimators, such as histograms and sampling methods, struggle to capture the correlations between multiple tables. On the other hand, current learning-based methods still suffer from the feature extraction of complex predicates and join relations, which will lead to inaccurate cost estimation, eventually a sub-optimal execution plan. To address these challenges, we present a novel end-to-end architecture leveraging deep learning to provide high-quality cardinality estimation. We exploit an effective feature extraction technique, which can fully make use of the structure of tables, join conditions and predicates. Besides, we use sampling-based technique to construct sample bitmaps for the tables and join conditions respectively. We also utilize the characteristics of predicate tree combined with recursive neural network to extract deep-level features of complex predicates. Finally, we embed these feature vectors into the model, which consists of three components: a recursive neural network, a graph convolutional neural network (GCN) and a multi-set convolutional neural network, to obtain the estimated cardinality. Extensive results conducted on real-world workloads demonstrate that our approach can achieve significant improvement in accuracy and be extended to queries with complex semantics.

A Contract-aware and Cost-effective LSM Store for Cloud Storage with Low Latency Spikes

Cloud storage is gaining popularity because features such as pay-as-you-go significantly reduce storage costs. However, the community has not sufficiently explored its contract model and latency characteristics. As LSM-Tree-based key-value stores (LSM stores) become the building block for numerous cloud applications, how cloud storage would impact the performance of key-value accesses is vital. This study reveals the significant latency variances of Amazon Elastic Block Store (EBS) under various I/O pressures, which challenges LSM store read performance on cloud storage. To reduce the corresponding tail latency, we propose Calcspar, a contract-aware LSM store for cloud storage, which efficiently addresses the challenges by regulating the rate of I/O requests to cloud storage and absorbing surplus I/O requests with the data cache. We specifically developed a fluctuation-aware cache to lower the high latency brought on by workload fluctuations. Additionally, we build a congestion-aware IOPS allocator to reduce the impact of LSM store internal operations on read latency. We evaluated Calcspar on EBS with different real-world workloads and compared it to the cutting-edge LSM stores. The results show that Calcspar can significantly reduce tail latency while maintaining regular read and write performance, keeping the 99 th percentile latency under 550μs and reducing average latency by 66%. In addition, Calcspar has lower write prices and average latency compared to Cloud NoSQL services offered by cloud vendors.

To store or not: Online cost optimization for running big data jobs on the cloud

As businesses increasingly rely on cloud-based big data analytics services to drive insights, reducing the cost of storing and analyzing large volumes of data in the cloud has become a major concern. During the execution of big data analysis jobs, some of the generated data can be reused by subsequent jobs. By storing such intermediate data, the cost of running big data jobs can be greatly reduced for businesses using cloud services. An important challenge is how to determine which data should be stored in order to save costs. Existing storing strategies do not differentiate between data with different usage frequencies, resulting in significant storage costs in practical applications. To address the aforementioned challenges, in this paper we propose two online algorithms, one deterministic and the other randomized, which dynamically determine whether to store the data with the aim of saving cost. We show that our proposed deterministic algorithm (resp., randomized) incurs costs within a factor of 2−α′ (resp., 21+α′) times the minimum cost obtained by an optimal offline algorithm which is assumed to know the exact future a priori. Finally, through extensive experiments with real-world workload of big data jobs in Alibaba Cloud environment, we demonstrate that our proposed online algorithms can achieve significant cost savings under common cloud pricing schemes.

Accelerating multi-tier storage cache simulations using knee detection

Storage cache hierarchies include diverse topologies, assorted parameters and policies, and devices with varied performance characteristics. Simulation enables efficient exploration of their configuration space while avoiding expensive physical experiments. Miss Ratio Curves (MRCs) efficiently characterize the performance of a cache over a range of cache sizes, revealing “key points” for cache simulation, such as knees in the curve that immediately follow sharp cliffs. Unfortunately, there are no automated techniques for efficiently finding key points in MRCs, and the cross-application of existing knee-detection algorithms yields inaccurate results.We present a multi-stage framework that identifies key points in any MRC, for both stack-based (e.g., LRU) and more sophisticated eviction algorithms (e.g., ARC). Our approach quickly locates candidates using efficient hash-based sampling, curve simplification, knee detection, and novel post-processing filters. We introduce Z-Method, a new multi-knee detection algorithm that employs statistical outlier detection to choose promising points robustly and efficiently.We evaluated our framework against seven other knee-detection algorithms, identifying key points in multi-tier MRCs with both ARC and LRU policies for 106 diverse real-world workloads. Compared to naïve approaches, our framework reduced the total number of points needed to accurately identify the best two-tier cache hierarchies by an average factor of approximately 5.5× for ARC and 7.7× for LRU.We also show how our framework can be used to seed the initial population for evolutionary algorithms. We ran 32,616 experiments requiring over three million cache simulations, on 151 samples, from three datasets, using a diverse set of population initialization techniques, evolutionary algorithms, knee-detection algorithms, cache replacement algorithms, and stopping criteria. Our results showed an overall acceleration rate of 34% across all configurations.

Data Placement Using a Classifier for SLC/QLC Hybrid SSDs

In hybrid SSDs (solid-state drives) consisting of SLC (single-level cell) and QLC (quad-level cell), efficiently using the limited SLC cache space is crucial. In this paper, we present a practical data placement scheme, which determines the placement location of incoming write requests using a lightweight machine-learning model. It leverages information about I/O workload characteristics and SSD status to identify cold data that does not need to be stored in the SLC cache with high accuracy. By strategically bypassing the SLC cache for cold data, our scheme significantly reduces unnecessary data movements between the SLC and QLC regions, improving the overall efficiency of the SSD. Through simulation-based studies using real-world workloads, we demonstrate that our scheme outperforms existing approaches by up to 44%.

Serverless Data Processing System and its Design Space Consideration

Serverless computing is becoming increasingly important in data-processing applications in science and business. The scheduler is at the centre of serverless data-processing systems, allowing for dynamic decisions on job and data placement. The complex design space, which is influenced by various user, cluster, and workload variables, presents problems for developing high-performance and cost-effective scheduling structures and processes. To make this exploration easier, we present Sched-Probe, a framework that includes a conceptual model and simulator for systematic design space exploration. Using the Sched-Probe framework, we evaluate the performance of three scheduling systems and two techniques using real-world workloads. Our open-source software is now available on ExDe, allowing system designers to collaborate on delving into the complexity of serverless scheduling, paving the way for optimised and efficient data-processing systems.

SafeCool: Safe and Energy-Efficient Cooling Management in Data Centers With Model-Based Reinforcement Learning

Optimizing the cooling system plays a central role for capping the data center power consumption. However, the performance of traditional cooling management strategies is not satisfactory due to the complexity of thermodynamic process. Recently, several works leveraged Reinforcement Learning (RL) to improve the energy efficiency of data center cooling system. While they demonstrated that it is possible to reduce the cooling power consumption via RL, there are still some key challenges that have to be addressed before field deployment, such as safe operation guarantee and sample complexity, etc. In this paper, we propose <monospace xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">SafeCool</monospace> , an actor-critic Model-Based Reinforcement Learning (MBRL) algorithm for data center cooling management. <monospace xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">SafeCool</monospace> incorporates two system models, i.e., a transition model to predict the future system state and a risk model to estimate the negative effect of executing an action. The safety of proposed algorithm is ensured by Model Predictive Control (MPC) and risk-guided exploration. In addition, by employing the MBRL framework, <monospace xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">SafeCool</monospace> achieves higher sample efficiency and accelerated convergence. Simulations using real-world workload trace reveal that <monospace xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">SafeCool</monospace> saves up to 13.18% cooling power compared with state-of-the-art MBRL data center cooling solutions.

PcGC: A Parity-Check Garbage Collection for Boosting 3-D NAND Flash Performance

Garbage collection or GC running in the controller of 3D NAND flash-based solid-state disks -SSDs -plays a critical role in the performance of storage systems. SSD manufacturers have developed various garbage collection solutions based on internal data movement or IDM to mitigate the impacts of garbage collection on request latency. Due to the circuit characteristics of flash memory, existing IDM-based garbage collection strategies are restricted by the page parity during data movement: odd pages must be migrated to odd pages, and even pages to even pages. When migrating two consecutive pages with the same parity, the free page between the two migrated pages will be wasted after the migration is complete. This ever-increasing page waste problem inevitably deteriorates the storage space utilization of flash memory, thereby degrading the overall performance of 3D NAND flash-based SSDs. To address this issue, we propose a parity-check garbage collection scheme called PcGC to revamp SSD performance by alleviating page waste during garbage collection. We build a parity-check unit in PcGC to facilitate checking the parity of migrated valid pages and destination pages. According to the parity results offered by the parity-check unit, PcGC dynamically adjusts the migration order of valid pages during the course of garbage collection. In doing so, PcGC fundamentally averts page waste caused by the page parity restriction, thereby enhancing 3D NAND flash performance. We quantitatively evaluate the performance of PcGC in terms of wasted pages, storage utilization, GC counts, write amplification, and average response time. We compare PcGC against the two state-of-the-art schemes -Amphibian and TTflash. The experimental results derived from the nine real-world workload traces unfold that compared with Amphibian and TTflash, (1) PcGC curtails the number of wasted pages by up to 91.4% with an average of 53.75%, (2) cuts back the number of GC counts by up to 52.2% with an average of 11.9%, and (3) slashes average write response time by up to 77.8% with an average of 13.0%.

ExDe: Design space exploration of scheduler architectures and mechanisms for serverless data-processing

Serverless computing is increasingly used for data-processing applications in both science and business domains. At the core of serverless data-processing systems is the scheduler, which ensures dynamic decisions about task and data placement. Due to the variety of user, cluster, and workload properties, the design space for high-performance and cost-effective scheduling architectures and mechanisms is vast. The large design space is difficult to explore and characterize. To help the system designer disentangle this complexity, we present ExDe, a framework to systematically explore the design space of scheduling architectures and mechanisms. The framework includes a conceptual model and a simulator to assist in design space exploration.We use the framework, and real-world workloads, to characterize the performance of three scheduling architectures and two mechanisms. Our framework is open-source software available on Zenodo.

A High-performance RDMA-oriented Learned Key-value Store for Disaggregated Memory Systems

Disaggregated memory systems separate monolithic servers into different components, including compute and memory nodes, to enjoy the benefits of high resource utilization, flexible hardware scalability, and efficient data sharing. By exploiting the high-performance RDMA (Remote Direct Memory Access), the compute nodes directly access the remote memory pool without involving remote CPUs. Hence, the ordered key-value (KV) stores (e.g., B-trees and learned indexes) keep all data sorted to provide range query services via the high-performance network. However, existing ordered KVs fail to work well on the disaggregated memory systems, due to either consuming multiple network roundtrips to search the remote data or heavily relying on the memory nodes equipped with insufficient computing resources to process data modifications. In this article, we propose a scalable RDMA-oriented KV store with learned indexes, called ROLEX, to coalesce the ordered KV store in the disaggregated systems for efficient data storage and retrieval. ROLEX leverages a retraining-decoupled learned index scheme to dissociate the model retraining from data modification operations via adding a bias and some data movement constraints to learned models. Based on the operation decoupling, data modifications are directly executed in compute nodes via one-sided RDMA verbs with high scalability. The model retraining is hence removed from the critical path of data modification and asynchronously executed in memory nodes by using dedicated computing resources. ROLEX efficiently alleviates the fragmentation and garbage collection issues, due to allocating and reclaiming space via fixed-size leaves that are accessed via the atomic-size leaf numbers. Our experimental results on YCSB and real-world workloads demonstrate that ROLEX achieves competitive performance on the static workloads, as well as significantly improving the performance on dynamic workloads by up to 2.2× over state-of-the-art schemes on the disaggregated memory systems. We have released the open-source codes for public use in GitHub.

LDPC Level Prediction Toward Read Performance of High-Density Flash Memories

High-density NAND flash memories have been prevailing in storage systems to achieve large capacities for explosive data. However, they suffer from more severe reliability degradation due to the narrowed margins between threshold voltage states. LDPC codes have been widely applied in high-density flash memories to ensure data reliability. Due to the increased number of cell states, more read voltages are required in reading a flash page correctly. This induces more soft levels to read pages with high bit error rates in LDPC decoding. Read latency is significantly increased in high-density flash memories. To enhance the read performance of high-density flash memories, this paper proposes PreLDPC, an LDPC level prediction approach with fine-grained LDPC reading. The key idea of PreLDPC is to predict the final read level during the early read iteration, thus avoiding unnecessary read-retry latency. From a preliminary study, we observe that after decoding in the first two iterations, the ratio of cells that lie in error-prone area (i.e. adjacent area of two cell states) can be obtained. The ratio is closely related to the final read level for a successful decoding. By exploiting this observation, PreLDPC directly uses the predicted read level for LDPC reading, which could eliminate excessive number of read retries. Furthermore, by exploiting the benefit of fine-grained LDPC reading, this paper further divides the existing integer level (called i-level, e.g. level-1 and level-2) into finer decimal level (called d-level, e.g. level-1.25 and level-1.5), and proposes a fine-grained read method. By combining the prediction method and fine-grained method together, PreLDPC can first estimate the i-level and then perform the read-retry iteration with d-levels to eliminate unnecessary read latency as much as possible. From experimental results of real-world workloads on Disksim with SSD extensions, it is verified that PreLDPC can effectively reduce read latency in high-density flash memories.

EAEFA: An Efficient Energy-Aware Task Scheduling in Cloud Environment

The scheduling of tasks in the cloud is a major challenge for improving resource availability and decreasing the total execution time and energy consumption of operations. Due to its simplicity, efficiency, and effectiveness in identifying global optimums, electric fish optimisation (EFO) has recently garnered a lot of interest as a metaheuristic method for solving optimisation issues. In this study, we apply electric fish optimisation (EAEFA) to the problem of cloud task scheduling in an effort to cut down on power usage and turnaround time. The objective is to finish all tasks in the shortest possible time, or makespan, taking into account constraints like resource availability and task dependencies. In the EAEFA approach, a school of electric fish is used to solve a multi-objective optimization problem that represents the scheduling of tasks. Because electric fish are drawn to high-quality solutions and repelled by low-quality ones, the algorithm is able to converge to a global optimum. Experiments validate EAEFA's ability to solve the task scheduling issue in cloud computing. The suggested scheduling strategy was tested on HPC2N and other large-scale simulations of real-world workloads to measure its makespan time, energy efficiency and other performance metrics. Experimental results demonstrate that the proposed EAEFA method improves performance by more than 30% with respect to makespan time and more than 20% with respect to overall energy consumption compared to state-of-the-art methods.

SplitPlace: AI Augmented Splitting and Placement of Large-Scale Neural Networks in Mobile Edge Environments

In recent years, deep learning models have become ubiquitous in industry and academia alike. Deep neural networks can solve some of the most complex pattern-recognition problems today, but come with the price of massive compute and memory requirements. This makes the problem of deploying such large-scale neural networks challenging in resource-constrained mobile edge computing platforms. To solve this, a promising solution is to split resource-hungry neural networks into lightweight disjoint smaller components for pipelined distributed processing. At present, there are two main approaches to do this: semantic and layer-wise splitting. The former partitions a neural network into parallel disjoint models that produce a part of the result, whereas the latter partitions into sequential models that produce intermediate results. However, there is no intelligent algorithm that decides which splitting strategy to use and places such modular splits to edge nodes for optimal performance. To combat this, this work proposes a novel AI-driven online policy, SplitPlace, that uses Multi-Armed-Bandits to intelligently decide between layer and semantic splitting strategies based on the input task&#x0027;s service deadline demands. SplitPlace places such neural network split fragments on mobile edge devices using decision-aware reinforcement learning for efficient and scalable computing. Moreover, SplitPlace fine-tunes its placement engine to adapt to volatile environments. Our experiments on physical mobile-edge environments with real-world workloads show that SplitPlace can significantly improve the state-of-the-art in terms of average response time, deadline violation rate, inference accuracy, and total reward by up to 46, 69, 3 and 12 percent respectively.