Evaluating multi-agent system architectures A case study concerning dynamic resource alloca

合集下载

外文翻译外文文献英文文献国际建设工程风险分析

外文翻译外文文献英文文献国际建设工程风险分析

外文文献:This analysis used a case study methodology to analyze the issues surrounding the partial collapse of the roof of a building housing the headquarters of the Standards Association of Zimbabwe (SAZ). In particular, it examined the prior roles played by the team of construction professionals. The analysis revealed that the SAZ’s traditional construction project was generally characterized by high risk. There was a clear indication of the failure of a contractor and architects in preventing and/or mitigating potential construction problems as alleged by the plaintiff. It was reasonable to conclude that between them the defects should have been detected earlier and rectified in good time before the partial roof failure. It appeared justified for the plaintiff to have brought a negligence claim against both the contractor and the architects. The risk analysis facilitated, through its multi-dimensional approach to a critical examination of a construction problem, the identification of an effective risk management strategy for future construction prject and riskThe structural design of the reinforced concrete elements was done by consulting engineers Knight Piesold (KP). Quantity surveying services were provided by Hawkins, Leshnick & Bath (HLB). The contract was awarded to Central African Building Corporation (CABCO) who was also responsible for the provision of a specialist roof structure using patented “gang nail” roof trusses. The building construction proceeded to completion and was handed over to the owners on Sept. 12, 1991. The SAZ took effective occupation of the headquarters building without a certificate of occupation. Also, the defects liability period was only three months .The roof structure was in place 10 years At first the SAZ decided to go to arbitration, but this failed to yield an immediate solution. The SAZ then decided toproceed to litigate in court and to bring a negligence claim against CABCO. The preparation for arbitration was reused for litigation. The SAZ’s quantified losses stood at approximately $ 6 million in Zimbabwe dollars (US $1.2m) .After all parties had examined the facts and evidence before them, it became clear that there was a great probability that the courts might rule that both the architects and the contractor were lia ble. It was at this stage that the defendants’ lawyers requested that the matter be settled out of court. The plaintiff agreed to this suxamined the prior roles played by the project management function and construction professionals in preventing/mitigating potential construction problems. It further assessed the extent to which the employer/client and parties to a construction contract are able to recover damages under that contract. The main objective of this critical analysis was to identify an effective risk management strategy for future construction projects. The importance of this study is its multidimensional examination approach.Experience sugge be misleading. All construction projects are prototypes to some extent and imply change. Change in the construction industry itself suggests that past experience is unlikely to be sufficient on its own. A structured approach is required. Such a structure can not and must not replace the experience and expertise of the participant. Rather, it brings additional benefits that assist to clarify objectives, identify the nature of the uncertainties, introduces effective communication systems, improves decision-making, introduces effective risk control measures, protects the project objectives and provides knowledge of the risk history .Construction professionals need to know how to balance the contingencies of risk with their specific contractual, financial, operational and organizational requirements. Many construction professionals look at risks in dividually with a myopic lens and donot realize the potential impact that other associated risks may have on their business operations. Using a holistic risk management approach will enable a firm to identify all of the organization’s business risks. This will increas e the probability of risk mitigation, with the ultimate goal of total risk elimination .Recommended key construction and risk management strategies for future construction projects have been considered and their explanation follows. J.W. Hinchey stated th at there is and can be no ‘best practice’ standard for risk allocation on a high-profile project or for that matter, any project. He said, instead, successful risk management is a mind-set and a process. According to Hinchey, the ideal mind-set is for the parties and their representatives to, first, be intentional about identifying project risks and then to proceed to develop a systematic and comprehensive process for avoiding, mitigat and its location. This is said to be necessary not only to allow alternative responses to be explored. But also to ensure that the right questions are asked and the major risks identified. Heads of sources of risk are said to be a convenient way of providing a structure for identifying risks to completion of a participant’s pa rt of the project. Effective risk management is said to require a multi-disciplinary approach. Inevitably risk management requires examination of engineering, legal and insurance related solutions .It is stated that the use of analytical techniques based on a statistical approach could be of enormous use in decision making . Many of these techniques are said to be relevant to estimation of the consequences of risk events, and not how allocation of risk is to be achieved. In addition, at the present stage of the development of risk management, Atkinson states that it must be recognized that major decisions will be made that can not be based solely on mathematical analysis. The complexity ofconstruction projects means that the project definition in terms of both physical form and organizational structure will be based on consideration of only a relatively small number of risks . This is said to then allow a general structured approach that can be applied to any construction project to increase the awareness of participants .The new, simplified Construction Design and Management Regulations (CDM Regulations) which came in to f 1996, into a single regulatory package.The new CDM regulations offer an opportunity for a step change in health and safety performance and are used to reemphasize the health, safety and broader business benefits of a well-managed and co-ordinated approach to the management of health and safety in construction. I believe that the development of these skills is imperative to provide the client with the most effective services available, delivering the best value project possible.Construction Management at Risk (CM at Risk), similar to established private sector methods of construction contracting, is gaining popularity in the public sector. It is a process that allows a client to select a construction manager (CM) based on qualifications; make the CM a member of a collaborative project team; centralize responsibility for construction under a single contract; obtain a bonded guaranteed maximum price; produce a more manageable, predictable project; save time and money; and reduce risk for the client, the architect and the CM.CM at Risk, a more professional approach to construction, is taking its place along with design-build, bridging and the more traditional process of design-bid-build as an established method of project delivery.The AE can review to get the projec. Competition in the community is more equitable: all subcontractors have a fair shot at the work .A contingency within the GMP covers unexpected but justifiable costs, and a contingency above the GMP allows for client changes. As long as the subcontractors are within the GMP they are reimbursed to the CM, so the CM represents the client in negotiating inevitable changes with subcontractors.There can be similar problems where each party in a project is separately insured. For this reason a move towards project insurance is recommended. The traditional approach reinforces adversarial attitudes, and even provides incentives for people to overlook or conceal risks in an attempt to avoid or transfer responsibility.A contingency within the GMP covers unexpected but justifiable costs, and a contingency above the GMP allows for client changes. As long as the subcontractors are within the GMP they are reimbursed to the CM, so the CM represents the client in negotiating inevitable changes with subcontractors.There can be similar problems where each party in a project is separately insured. For this reason a move towards project insurance is recommended. The traditional approach reinforces adversarial attitudes, and even provides incentives for people to overlook or conceal risks in an attempt to avoid or transfer responsibility.It was reasonable to assume that between them the defects should have been detected earlier and rectified in good time before the partial roof failure. It did appear justified for the plaintiff to have brought a negligence claim against both the contractor and the architects.In many projects clients do not understand the importance of their role in facilitating cooperation and coordination; the desi recompense. They do not want surprises, and are more likely to engage in litigation when things go wrong.中文译文:国际建设工程风险分析索赔看来是合乎情理的。

体系结构设计英语

体系结构设计英语

体系结构设计英语Designing System ArchitecturesThe field of system architecture has become increasingly crucial in the modern technology landscape. As the complexity of software and hardware systems continues to grow, the need for robust and efficient architectural designs has become paramount. System architecture encompasses the fundamental organization and structure of a system, including its components, their interactions, and the principles that govern its design and evolution.One of the key aspects of system architecture is the ability to decompose a complex system into manageable and well-defined subsystems. This modular approach allows for better scalability, maintainability, and flexibility, as changes can be made to individual components without disrupting the entire system. By identifying the appropriate level of abstraction and defining clear interfaces between subsystems, system architects can create architectures that are easier to understand, develop, and evolve over time.Another crucial aspect of system architecture is the consideration of non-functional requirements such as performance, security, reliability,and scalability. These requirements often have a significant impact on the overall design and can influence the selection of specific architectural patterns and technologies. For example, a system designed for high-performance computing may prioritize parallel processing and distributed computing, while a system focused on secure data storage may emphasize encryption and access control mechanisms.The process of designing a system architecture typically involves several key steps. First, the system's requirements and goals must be clearly defined, taking into account both functional and non-functional requirements. This often involves a thorough analysis of the problem domain, stakeholder needs, and any existing constraints or limitations.Next, the system architect must identify the appropriate architectural styles and patterns that can effectively address the identified requirements. This may involve exploring various architectural approaches, such as client-server, microservices, event-driven, or n-tier architectures, and evaluating their suitability for the specific system being designed.Once the architectural style has been selected, the system architect must define the high-level components and their interactions. This may involve creating detailed component diagrams, data flowdiagrams, and other visual representations to ensure a clear understanding of the system's structure and behavior.As the design process progresses, the system architect must also consider the deployment and operational aspects of the system. This may include defining the infrastructure requirements, such as hardware, software, and network configurations, as well as the processes and tools needed for deployment, monitoring, and maintenance.Throughout the design process, the system architect must also engage in ongoing communication and collaboration with various stakeholders, including developers, project managers, and end-users. This ensures that the system architecture remains aligned with the evolving needs and constraints of the project, and that any changes or refinements can be effectively incorporated into the design.In conclusion, system architecture is a critical discipline that underpins the success of complex software and hardware systems. By leveraging modular design, considering non-functional requirements, and engaging in a structured design process, system architects can create architectures that are scalable, maintainable, and adaptable to the ever-changing technological landscape. As the demand for sophisticated and reliable systems continues to grow, the role of the system architect will become increasingly essential in drivinginnovation and ensuring the long-term success of technology-driven organizations.。

陆战平台分布式综合模块化系统架构建模方法

陆战平台分布式综合模块化系统架构建模方法

收稿日期:2020-02-08修回日期:2020-03-18作者简介:王昊(1996-),男,山西平遥人,硕士研究生。

研究方向:系统工程。

摘要:分布式综合模块化系统架构,成为航空应用领域的主流架构和发展趋势。

我国空军已进行了分布式综合模块化系统架构研究设计和仿真评估,但缺乏系统级的架构优劣评估。

兵器领域已将综合模块化系统架构应用于型号项目中,但在分布式综合模块化架构方面,尚未在具体项目中应用。

针对分布式综合模块化系统架构缺乏系统级评估手段的问题,提出一种陆战平台分布式综合模块化系统架构建模方法。

根据系统架构评估需求,建立不同层级实体的模型及约束,分别对应陆战平台信息控制系统的业务层、操作系统分区层、模块层、子系统层、系统层;在层级划分的基础上,定义各层级模型的属性;通过对评估指标中的综合度、耦合度进行评估计算,对系统架构建模方法进行验证。

系统架构建模方法为系统架构评估提供了一种参考。

关键词:系统架构,建模方法,陆战平台,综合模块化中图分类号:TJ811;TP311文献标识码:ADOI :10.3969/j.issn.1002-0640.2021.03.016引用格式:王昊,张振华,赵刚,等.陆战平台分布式综合模块化系统架构建模方法[J ].火力与指挥控制,2021,46(3):92-99.陆战平台分布式综合模块化系统架构建模方法王昊,张振华,赵刚,梁栋,贾智(北方自动控制技术研究所,太原030012)Research on Modeling Method of Distributed Integrated ModularSystem Architecture of Land Combat PlatformWANG Hao ,ZHANG Zhen-hua ,ZHAO Gang ,LIANG Dong ,JIA Zhi (North Automatic Control Technology Institute ,Taiyuan 030012,China )Abstract :The distributed integrated modular system architecture has become the mainstreamarchitecture and development trend in the aviation application field.The Chinese Air Force has conducted research and design and simulation evaluation of the distributed integrated modular system architecture ,but it lacks a system -level evaluation of the advantages and disadvantages of the architecture.In the field of weapons ,the integrated modular system architecture has been applied to model projects ,but the distributed integrated modular architecture has not yet been applied to specific projects.Aiming at the problem of the lack of system -level evaluation methods for the distributed integrated modular system architecture ,this paper proposes a modeling method for the distributed integrated modular system architecture of the land warfare platform.According to the evaluation requirements of the system architecture ,this paper establishes the models and constraints of different levels of entities ,corresponding to the business layer ,operating system partition layer ,module layer ,subsystem layer ,and system layer of the land warfare platform information control system ;on the basis of layer division ,Define the attributes of each level model ;verify the system architecture modeling method by evaluating the degree of integration and coupling in the evaluation indicators.The system architecture modeling method in this article provides a reference for system architecture evaluation.Key words :system structure ,modeling method ,land combat platform ,integrated modular Citation format :WANG H ,ZHANG Z H ,ZHAO G ,et al.Research on modeling method of distribut-ed integrated modular system architecture of land combat platform [J ].Fire Control &Command Control ,2021,46(3):92-99.文章编号:1002-0640(2021)03-0092-08Vol.46,No.3Mar ,2021火力与指挥控制Fire Control &Command Control 第46卷第3期2021年3月92··(总第46-)0引言系统架构可以拆分成两部分:“系统”和“架构”。

Multi-agent System中多维度信誉模型设计

Multi-agent System中多维度信誉模型设计
地选 择交互 对象从 而最大化 交 互 的 收益 l 3 ] 具 有重 要
的理论 和 实践意 义.
Ab s t r a c t : Re p u t a t i o n me c h a n i s m wa s i n t r o d u c e d t o s o l v e c o mp l e x i n t e r a c t i o n p r o b l e ms a n d p r o mo t e c o o p e r a t i o n i n mu l t i — a g e n t s y s t e m (M A S) . B a s e d o n t wo r e p u t a t i o n
文 献标 志码 : A 组 织 系统 的研 究 与实 现l 】 ] . MAS中 Ag e n t 的交互 是

个充 斥 着 大 量 不 确 定 性 和 不 完 全 信 息 的 复 杂 过
Mu l t i - d i me ns i o n Re p u t a t i o n Mo d e l De s i g n i n 程. 信誉 是 简化复 杂 系统 的有效 机制 , 可 以成 为解 决 Mu l t i — a g e n t S y s t e m Ag e n t 交互选 择 问题 的一种 简单 的方法 . Ag e n t 在交
Ma r .2 0 1 3
文章编号 : 0 2 5 3 — 3 7 4 X( 2 0 1 3 ) 0 3 — 0 4 7 6 — 0 7
Mu l t i . a g e n t S y s t e m 中 多维 度 信 誉 模 型 设 计
徐 莉, 余 红伟
( 武汉大学 经济与管理学院 , 湖北 武汉 4 3 0 0 7 2 )

FortiManager 自动化驱动中心化管理系统说明书

FortiManager 自动化驱动中心化管理系统说明书

DATA SHEETFortiManagerAutomation-Driven Centralized Management Manage all your Fortinet devices in a single-console central management system. FortiManager provides full visibility of your network, offering streamlined provisioning and innovative automation tools.Integrated with Fortinet’s Security Fabric , the security architecture and FortiManager’s Automation Driven Network Operations capabilities provide a foundation to secure and optimize network security , such as provisioning and monitoring SD-WANs.Orchestrate security devices and systems on-premise or in the cloud to streamline network provisioning, security policy updates & change management.Automate your time-intensive processes and accelerate workflows to offload NOC-SOC, reduce administrative tasks and address talent shortages.Optimize Visibility to the entire digital attack surface and awareness of increasing cyber threats from one centralized location, through accurate detection, automated correlation and rapid response features.§ § § § § § §DATA SHEET | FortiManager2HighlightsSingle Pane Automation and OrchestrationFortinet Security Fabric delivers sophisticated security management for unified, end-to-end protection. Deploying Fortinet-based security infrastructure to battle advanced threats, and adding FortiManager to provide single-pane-of-glass management across your entire extended enterprise provides insight into network-wide traffic and threats.FortiManager offers enterprise-class features to contain advanced threats. FortiManager also delivers the industry’s best scalability to manage up to 100,000 Fortinet devices. FortiManager, coupled with the FortiAnalyzer family of centralized logging and reporting appliances,provides a comprehensive and powerful centralized management solution for your organization.Centralized SD-WAN Deployment & MonitoringPowerful SD-WAN management capabilities by using templates. Enhanced SD-WAN monitoring for each SD-WAN link member with visibility of link status, application performance, bandwidth utilization. The SLA targets are included in performance monitoring graphs for each WAN provider.Configuration and Settings ManagementCollectively configure the device settings - using the provisioning templates and advance CLI templates improves management of a large number of devices. Automatic device configuration backup with revision control and change audit make it easier for daily administrative tasks.Central Management of Network InfrastructureCentrally manage FortiGate , FortiSwitch, FortiExtender, FortiAP . The VPN manager simplifies the deployment and allows centrally-provisioned VPN community and monitoring of VPN connections on Google Map. FortiAP Manager allows configuring, deploying and monitoring FortiAPs from a single console with Google Map view. The FortiClient Manager allows centralized configuration, deployment and monitoring of FortiClients.Multi-Tenancy & Role Based AdministrationFortiManager equips admins with granular device and role based administration for deploying multi-tenancy architecture to large enterprises, with a hierarchical objects database to facilitate re-use of common configurations and serve multiple customers. The graphical interface makes it easy to view, create, clone and manage ADOMs. You can use ADOMs to manage independent security environments, each ADOM with its own security policies and configuration database. FortiManager enables you to group devices logically or geographically for flexible management, and the zero-touch deployment uses templates to provision devices for quick mass deployment and supports firmware version enforcement. Define global objects such as Firewall Objects, Policies and Security Profiles to share across multiple ADOMs. Granular permissions allow assigning ADOMs, devices and policies to users based on role and duties.API for Automation and OrchestrationRESTful API allows MSSPs/large enterprises to create customized, branded web portals for policy and object administration. Automate common tasks such as provisioning new FortiGates and configuring existing devices. Join Fortinet Developer Network (FNDN) to access exclusive articles, how-to content for automation and customization, community-built tools, scripts and sample code.Security Policy ManagementA set of commonly used security policies can be now grouped in a Policy Block and inserted as needed in different Policy Packages.Global policy feature that allows companies such as: Telecom, MSSP , SAAS providers applies a header and/or footer policy at the ADOM level to all the policy packages or to a selection of packages, as needed.DATA SHEET | FortiManagerHighlightsFortiManager VMFortinet offers the FortiManager VM in a stackable license model. This model allows you to expand your VM solution as your environment expands. Utilizing virtualization technology, FortiManager-VM is a software-based version of the FortiManager hardware appliance and is designed to run on many virtualization platforms. It offers all the features of the FortiManager hardware appliance.The FortiManager virtual appliance family minimizes the effort required to monitor and maintain acceptable use policies, as well as identify attack patterns that can be used to fine tune the security policy, thwarting future attackers.SpecificationsFMG-VM-10-UG FMG-VM-100-UG FMG-VM-1000-UG FMG-VM-5000-UG10 +100 +1,000 +5,000 +200 GB 1 TB 4 TB8 TB251025VMware ESX/ESXi 5.0/5.1/5.5/6.0/6.5/6.7, Microsoft Hyper-V 2008 R2/2012/2012 R2/2016, Citrix XenServer 6.0+ and Open SourceXen 4.1+, KVM on Redhat 6.5+ and Ubuntu 17.04, Nutanix AHV (AOS 5.10.5), Amazon Web Services (AWS), Microsoft Azure, GoogleCloud (GCP), Oracle Cloud Infrastructure (OCI), Alibaba Cloud (AliCloud)vCPU Support (Minimum / Maximum) 2 / UnlimitedNetwork Interface Support (Min / Max) 1 / 4Integration & Security FabricIntegration with ITSM to mitigate security events and applyconfiguration changes and policy updates. Seamless integrationwith FortiAnalyzer appliances provides in-depth discovery, analysis,prioritization and reporting of network security events. Create fabricconnectors to facilitate connections with third-party vendors viapxGrid , OCI, ESXi and others, to share and exchange data.FortiManager’s workflow for audit and compliance enables youto review, approve and audit policy changes from a central place,including automated processes to facilitate policy compliance,policy lifecycle management, and enforced workflow to reduce riskfor policy changes.Monitor and Report for Deep VisibilityAccess vital security and network statistics, as well as real-timemonitoring and integrated reporting provides visibility into networkand user activity. For more powerful analytics, combine with aFortiAnalyzer appliance for additional data mining and graphicalreporting capabilities.Network & Security Operations VisibilityAutomated data exchanges between security (SOC) workflows andoperational (NOC) workflows, creating a single, complete workflowthat not only saves time, but also provides the capacity to completeadditional incident response activities. FortiManager’s NOC-SOCdelivers advanced data visualization to help Analysts quicklyconnect dots and identify threats, simplifying how organizationsdeliver security and remediate breaches, data exfiltration, andcompromised hosts.DATA SHEET | FortiManager4Safety CertificationscUL, CB CE, BSMI, KC, UL/cUL, CB, GOST FCC Part 15 Class A, C-Tick, VCCI, CE, UL/cUL, CBSpecifications1 Each Virtual Domain (VDOM) operating on a physical or virtual device counts as one (1) licensed network device. Global Policies and high availability support available on all models* Optional redundant AC power supply, not includedDATA SHEET | FortiManager5FMG-2000EFMG-3000FSafety CertificationscUL, CBCE, BSMI, KC, UL/cUL, CB, GOSTcUL, CB, GOSTSpecifications1 Each Virtual Domain (VDOM) operating on a physical or virtual device counts as one (1) licensed network device Global Policies and high availability support available on all models. 4 + Indicates Device Add-On License availableDATA SHEET | FortiManagerOrder InformationProduct SKU DescriptionFortiManager FMG-200F Centralized management, log and analysis appliance — 2xRJ45 GE, 2xSFP, 8 TB storage, up to 30x Fortinet devices/virtual domains.FMG-300F Centralized management, log and analysis appliance — 4x GE RJ45, 2xSFP, 16 TB storage, up to 100x Fortinet devices/virtual domains.FMG-1000F Centralized management, log and analysis appliance — 2x RJ45 10G, 2x SFP+ slots, 32 TB storage, up to 1000x Fortinet devices/virtual domains.FMG-2000E Centralized management, log and analysis appliance — 4x GE RJ45, 2x 10 GE SFP+ slots, 36 TB storage, dual power supplies, manages up to 1,200Fortinet devices/virtual domains.FMG-3000F Centralized management, log and analysis appliance — 4x GE RJ45, 2x 10 GE SFP+ slots, 48 TB storage, dual power supplies, manages up to 4,000Fortinet devices/virtual domains.FMG-3700F Centralized management, log and analysis appliance — 2x10GbE SFP+, 2x1GbE RJ-45 slots, 240 TB storage, dual power supplies, manages up to 10,000Fortinet devices/virtual domains.FortiManager Device Upgrade FMG-DEV-100-UG FortiManager device upgrade license for adding 100 Fortinet devices/VDOMs (3000 series and above - hardware only)FortiManager VM Built-in Evaluation Built-in 15-day EVAL license, no activation required.Full Evaluation (60-days)EVAL license. License and activation required.FMG-VM-Base Base license for stackable FortiManager-VM. Manages up to 10 Fortinet devices/Virtual Domains, 1 GB/Day of Logs and 100 GB storage capacity. Designedfor all supported platforms.FMG-VM-10-UG Upgrade license for adding 10 Fortinet devices/Virtual Domains; allows for total of 2 GB/Day of Logs and 200 GB storage capacity.FMG-VM-100-UG Upgrade license for adding 100 Fortinet devices/Virtual Domains; allows for total of 5 GB/Day of Logs and 1 TB storage capacity.FMG-VM-1000-UG Upgrade license for adding 1,000 Fortinet devices/Virtual Domains; allows for total of 10 GB/Day of Logs and 4 TB storage capacity.FMG-VM-5000-UG Upgrade license for adding 5,000 Fortinet devices/Virtual Domains; allows for total of 25 GB/Day of Logs and 8 TB storage capacity.Additional FortiManager Items FC-10-FDN1-139-02-12 1 Year Subscription Renewal for 1 User to Fortinet Developer NetworkFC-10-FDN2-139-02-12 1 Year Subscription for Unlimited Users to Fortinet Developer NetworkFMG-SDNS License to operate FortiManager as a dedicated Secure DNS server appliance (3000 series and above – hardware only) Copyright © 2019 Fortinet, Inc. All rights reserved. Fortinet®, FortiGate®, FortiCare® and FortiGuard®, and certain other marks are registered trademarks of Fortinet, Inc., and other Fortinet names herein may also be registered and/or common law trademarks of Fortinet. All other product or company names may be trademarks of their respective owners. Performance and other metrics contained herein were attained in internal lab tests under ideal conditions, and actual performance and other results may vary. Network variables, different network environments and other conditions may affect performance results. Nothing herein represents any binding commitment by Fortinet, and Fortinet disclaims all warranties, whether express or implied, except to the extent Fortinet enters a binding written contract, signed by Fortinet’s General Counsel, with a purchaser that expressly warrants that the identified product will perform according to certain expressly-identified performance metrics and, in such event, only the specific performance metrics expressly identified in such binding written contract shall be binding on Fortinet. For absolute clarity, any such warranty will be limited to performance in the same ideal conditions as in Fortinet’s internal lab tests. Fortinet disclaims in full any covenants, representations, and guarantees pursuant hereto, whether express or implied. Fortinet reserves the right to change, modify, transfer, or otherwise revise this publication without notice, and the most current version of the publication shall be applicable. Fortinet disclaims in full any covenants, representations, and guarantees pursuant hereto, whether express or implied. Fortinet reserves the right to change, modify, transfer, or otherwise revise this publication without notice, and the most current version of the publication shall be applicable.FST-PROD-DS-FMG FMG-DAT-R47-201908。

有人机_无人机编队协同任务分配方法

有人机_无人机编队协同任务分配方法
无人机 (unmanned aerial vehicle , UAV) 系统具有隐身 性能好 ,自主能力强 ,可重复回收利用等特点 ,在现代战场 上取得了越来越广泛的应用 ,其发展也受到了各国的重视 。 而日益复杂的战场环境 、日益多样的作战样式和日益扩大 的作战范围 ,使无人机的自主性和智能性面临严峻的挑战 , 要求其能够在复杂多变的战场环境中实时快速的做出正确 的决策 。有人机/ 无人机编队协同作战可以充分发挥人类 智能在关键时刻的作用 ,弥补无人机智能性的不足 ,使无人 机的优势得到更充分的发挥 ,提高系统环境适应能力和整 体作战效能 ,这一领域受到国外研究机构和学者的普遍
agent 接收任务 ,并根据任务属性进行任务规划和航路规 划 ,并执行任务 ,它们通过机载传感器系统感知环境 ,响应 环境变化 ,根据环境信息决定是否进行任务和航路的重规 划 ,并在任务执行过程中接收管理 agent 控制命令 ,向管理 agent 回传任务 、状态信息 ,与其他 UAV agent 相互通信和 协调 ,在探 测 到 突 发 威 胁 时 , 向 管 理 agent 和 其 他 UAV agent通报威胁信息 ,共同完成系统任务 。有人机/ 无人机 编队协同任务系统中 ,管理 agent 具有完全的智能性和自 治性 ,是完全自治 agent (auto nomous agent) ,而 UAV agent 由于受有人机指挥控制 ,具有部分自治性 ,是半自治 agent ( semi2auto no mous agent ) , 整 个 编 队 构 成 多 智 能 体 系 统 ( multi2agent system , MAS) ,通过各 agent 之间的交互和协 同实现整个系统的任务 ,实际上是一个有限中央控制下的 分布式系统 。如果是多编队协同作战 ,则通过各编队有人 机之间的交互实现各编队之间的协同 。有人机/ 无人机编 队 MAS 协同任务分配体系结构如图 2 所示 。图中 , Ti ( i = 1 ,2 , …, M) 为管理 agent 分解所得任务集合 。

基于Multi—Agent的总承包工程项目供应链信息协同机制研究

基于Multi—Agent的总承包工程项目供应链信息协同机制研究

关 键 词 :总承 包 工程 ;Mu i aet系统 ; 黑板 模 型 ;协 同机 制 l — gn t 中图分类号 :T 33 :F4 P9 72 文献标 识码 :A 文章编号 :10 00—79 (0 2 7— 16— 3 6 5 2 1 )1 0 4 0
I or a i n Sy r e i e ha s fGe r lCo r ci n tuc i n nf m to ne g tc M c nim o ne a nt a tng Co s r to
图 1 黑 板 模 型 示 意 图
“ 板模 型 ” 是 一 种典 型而 流 行 的专 家 系 统 结 黑 构 模 式 ,最 早 由美 国 C rei an g e—Me o l n大 学 开 发 的 l HE R A A S Y—U系统 中创立 ,后 来 被 许 多 系 统 所 效仿 和采 用 。黑板 模 型 由知 识 源 、黑 板 和 监 控 机 制 三个 基 本 部分 构 成 ,如 图 1所 示 _ 。黑 板 是 共 享 的 问题 9 求 解 工作 空 间 ,包 含存 储 数 据 、传 递 信 息 与处 理 方 法 的动 态 数 据 库 。一 般 按 照 层 次 结 构 的 方 式 组 织 , 在 问题 求 解 过 程 中 ,知 识 源 之 间 的交 互 与 通 讯 通 过 黑 板 进行 。所 谓 知识 源是 指 一 个 知 识 模 块 。 黑 板 结 构 中具 有 多 个 知 识 源 ,每 一 知 识 源 由条 件 部 分 和 动 作 部分 组 成 ,可 以完 成 某 些 特定 的解 题 功 能 。条 件 旦满 足 ,知 识 源 就 会 触 发 ,其 动 作 部 分 增 加 或 修 改 黑板 的记 录 。 由监 督 程 序 和 调 度 程 序 组 成 的监 控 机 制是 黑板 模 型求 解 问题 的 推理 机 构 。监 督 程 序 时 刻 注视着 黑 板 状 态 ,根 据 黑 板 信 息 的状 态 变 化 ,监 督程 序采 用 某 种 策 略选 择 合 适 的知 识 源 ,将 其 条 件 部分 放人 调 度 队列 ,随 后 ,若 条 件 部 分 与 黑 板 状 态 匹 配成 功 ,则 将 其 动 作 部 分 放 人 调 度 队列 。而 动 作

浙江省二级注册建筑师初始注册流程

浙江省二级注册建筑师初始注册流程

浙江省二级注册建筑师初始注册流程1.首先,申请人需要登录浙江省建设工程信息管理系统,填写个人信息。

First, the applicant needs to log in to the Zhejiang Province Construction Engineering Information Management System and fill in personal information.2.接着,申请人需要准备相关的证明材料,包括身份证、学历证明、职业资格证书等。

Next, the applicant needs to prepare relevant supporting materials, including ID card, academic qualifications, professional qualifications, etc.3.申请人需要缴纳注册费用,并获取收据作为证明。

The applicant needs to pay the registration fee and obtain a receipt as proof.4.然后,申请人需要参加浙江省建筑师管理部门组织的初审。

Then, the applicant needs to participate in thepreliminary review organized by the Zhejiang Provincial Architectural Management Department.5.初审合格后,申请人将接受现场面试和考试。

After passing the preliminary review, the applicant will undergo on-site interviews and exams.6.面试和考试的内容包括建筑设计、规划调查、工程监理等方面的知识。

The interviews and exams cover knowledge in architectural design, planning and surveying, project supervision, etc.7.面试考试通过后,申请人将进行注册建筑师的执业注册审核。

  1. 1、下载文档前请自行甄别文档内容的完整性,平台不提供额外的编辑、内容补充、找答案等附加服务。
  2. 2、"仅部分预览"的文档,不可在线预览部分如存在完整性等问题,可反馈申请退款(可完整预览的文档不适用该条件!)。
  3. 3、如文档侵犯您的权益,请联系客服反馈,我们会尽快为您处理(人工客服工作时间:9:00-18:30)。

Evaluating Multi-Agent System Architectures:A case study concerning dynamic resource allocationPaul Davidsson Stefan JohanssonDepartment of Software Engineering and Computer Science,Blekinge Institute of Technology,Soft Center,37225Ronneby,Swedenpdv,sja@bth.se,Abstract.Much effort has been spent on suggesting and implementing new ar-chitectures of Multi-Agent Systems.However,we believe the time has come tocompare and evaluate these architectures in a more systematic way.Rather thanjust studying a particular application,we suggest that more general problem do-mains corresponding to sets of applications should be studied.Similarly,we ar-gue that it is more useful to study the properties of classes of multi-agent systemarchitectures than particular architectures.Also,it is important to evaluate the ar-chitectures in several dimensions,both different performance-related attributes,which are domain dependent and more general quality attributes,such as,ro-bustness,modifiability,and scalability.As a case study we investigate the generalproblem of”dynamic resource allocation”and present four classes of multi-agentsystem architectures that solve this problem.These classes are discriminated bytheir degree of distribution of control and degree of synchronization.Finally,weinstantiate each of these architecture classes and evaluate,through simulationexperiments,how they solve a concrete dynamic resource allocation problem,namely load balancing and overload control of Intelligent Networks.1IntroductionMuch effort has been spent on suggesting and implementing new architectures of Multi-Agent Systems(MAS).Unfortunately,this work has been carried out in a quite unstruc-tured way where a(group of)researcher(s)invents a new architecture and applies it to a particular domain and conclude that it seems to be appropriate for this domain.Often this new architecture is not even compared to any existing architecture.We believe that this area now has reached the level of maturity when it is appropriate to compare and evaluate these architectures in a more systematic way.1.1ApplicationsOf course,there is no single MAS architecture that is the most suitable for all appli-cations.On the other hand,tofind out whether one architecture performs better than another for a particular application is usually of limited scientific interest.(Although this information may be very useful to solve that particular problem.)Instead,we sug-gest the study of more general problem domains corresponding to sets of applicationswith common characteristics.In this paper we will exemplify this approach by inves-tigating the general problem of dynamic resource allocation.Such studies tend to be quite abstract and are for that reason often of a theoretical and qualitative nature.They therefore should be supplemented with and validated by quantitative empirical studies in one or more concrete applications corresponding to instances of the general domain, in this paper exemplified by load balancing and overload control in Intelligent Networks,a type of telecommunication system.1.2ArchitecturesJust as it is useful to study classes of applications rather than particular applications, we argue that it is useful to study classes of MAS architectures in addition to particular architectures.To make such studies possible,we need to describe MAS architectures in a way that abstracts the particularities of the individual architectures but still captures their relevant characteristics.To develop a general way of characterizing MAS architec-tures is in itself a major research task.In fact,it may be necessary to use several views to capture all relevant aspects of an architecture cf.Kruchten[9].In this work we will categorize MAS architectures according to two properties:the type of control used(from fully centralized to fully distributed),the type of coordi-nation(synchronous vs.asynchronous).As with classes of applications,it is mostly theoretical studies that can be performed on classes of MAS architectures.Therefore, they often need to be supplemented with empirical studies using instantiations of these architectures.Below we will present four concrete architectures corresponding to dif-ferent combinations of the two architectural properties.1.3Quality attributesIt is possible to evaluate MAS architectures with respect to several different quality attributes,both different performance-related attributes and more general quality at-tributes,such as,robustness,modifiability,and scalability.Some of these attributes are domain independent and some are specific for each set of applications,e.g.,performance-related attributes.As mentioned earlier,we do not think it is possible tofind a MAS ar-chitecture that is optimal with respect to all relevant attributes.Rather,there is an inher-ent trade-off between these attributes and different architectures balance this trade-off in various ways.The various applications,on the other hand,require different balances of this trade-off.Thus,in order to choose the right architecture for a particular applica-tion,knowledge about relevant attributes and how different MAS architectures support them is essential.1.4Evaluation frameworkTo evaluate a set of architectures in a systematic way,we suggest an approach that can be described in terms of the following three-dimensional space:–the set of possible applications,–the set of possible MAS architectures,and–the set of attributes used to evaluate the architectures.The suggested approach is to investigate substantial parts of this space rather than just single points.We believe that this approach,besides of enabling a more systematic investigation of the space,will lead to a deeper understanding of MAS s and their appli-cations,which,in turn,will contribute to reach the long-term goal of obtaining general design principles of MAS s.In this paper we will apply this approach to the general problem of dynamic resource allocation and present four abstract MAS architectures with different characteristics that solves the problem.These will then be compared with respect to a number attributes, e.g.,reactivity,ability to balance the loads,fairness,utilization of resources,respon-siveness,amount of communication overhead,robustness,modifiability,and scalability. Finally,we evaluate concrete instantiations of these abstract architectures in a concrete dynamic resource allocation problem,namely load balancing and overload control in Intelligent Networks.2Abstract domain:Dynamic resource allocationMulti-agent technology has proved to be successful for dynamic resource allocation, e.g.power load management[11]and cellular phone bandwidth allocation[3].Basi-cally,this problem concerns the allocating of resources between a number of customers, given a number of providers.The dynamics of the problem lies in that the needs of the customers,as well as the amount of resources made available by the providers,vary over time.The needs and available resources not only vary on an individual level,but also the total needs and available resources within the system.We will here assume that the resources cannot be buffered,i.e.,they have to be consumed immediately,and that the cost of communication(and transportation of resources)between any customer-provider pair is equal.2.1Abstract multi-agent architecturesThere are many ways of dividing the set of possible MAS architectures into different subsets based on their characteristics,e.g.:–the topography of the system,–the degree of mobility and dynamics of the communications,–the degree of distribution of control,and–the degree of synchronization of interaction.We have chosen to focus the two last properties.By degree of distribution we mean to what degree the control of the system is distributed.The degree of synchronization is a measure of how the execution of the agents interrelate with each other.We may have agents that are highly sophisticated,but who only interacts at special slots in time,and thus have a high degree of synchronization.There are also systems in which the agents may interact continuously,independently of when other agents interact,which we will refer to as asynchronous.To sum up,we will compare the following four abstract classes of MAS architec-tures for dynamic resource allocation:centralized synchronous architectures,central-ized asynchronous architectures,distributed synchronous architectures,and distributed asynchronous architectures.2.2Abstract attributesWe have identified the following important performance-related attributes to dynamic resource allocation:–Reactivity:How fast are resources re-allocated when there are changes in demand?–Load balancing:How evenly is the load balanced between the resource providers?–Fairness:Are the customers treated equally?–Utilization of resources:Are the available resources utilized as much as is possible?–Responsiveness:How long does it take for the customers to get response to a re-quest?–Communication overhead:How much extra communication is needed for the re-source allocation?In addition,there are a number of more general software architecture quality attributes [6]that should be addressed,e.g.:–Robustness:How vulnerable is the system to node or link failures?–Modifiability:How easy is it to change the system after it is implemented(and often deployed)?–Scalability:How good is the system at handling large numbers of users(providers and customers)?2.3Theoretical/qualitative evaluationWe will now make a brief theoretical,or qualitative,analysis of how the degree distribu-tion and synchronization of the multi-agent system architecture influence the attributes identified in the last section.–Reactivity should be promoted by asynchronous architectures since there is no need to await any synchronization event before i)an agent can notify other agents about changes in demand and ii)other agents can take the appropriate actions to adapt to these changes.–Load balancing should be favored,or at least not disfavored,by centralized control since it is possible to take advantage of the global view of the state of the system,e.g.,the current load at the providers and the current demand of the customers.–Similarly should fairness be easier to achieve for architectures with centralized con-trol since they have information about the global state of the system.–It is not clear from a strictly theoretical analysis if there is any correlation between the ability to utilize the resources and the architectural properties.Empirical studies are probably necessary.–Also,it is not clear from a strictly theoretical analysis if there is any correlation between responsiveness and the architecture properties.–Communication overhead can be measured either by the number of messages sent, or by the bandwidth required for the allocation.Synchronous architectures tend to concentrate the message sending to short time intervals,and thus requiring a large bandwidth,whereas asynchronous architectures tend to be better at utilizinga given bandwidth over the time.Also,communication in distributed architectureshas a tendency to be more local than in centralized architecture,using smaller parts of the network.–Regarding robustness we conclude that the more centralized the control is,the more vulnerable the system gets.Basically,the reason for this is that the system cannot function,i.e.,perform reallocation,if the agents that are responsible for the con-trol fail.In more distributed systems,the reallocation may function partially even though some agents have failed.–The modifiability,to add or remove a provider or customer,seems to be better in centralized architectures.For instance,changes may only be necessary in one part of the system.–Scalability seems to be better supported by distributed architectures than central-ized architectures.Firstly,the computational load for the resource allocation is di-vided between a number of computers,and secondly,the risk for communication bottlenecks is smaller.3Concrete domain:Load balancing in Intelligent NetworksOne important area in which the dynamic resource allocation problem is present is telecommunications.The Intelligent Network(IN)concept was developed in order to enable operators of telecommunication networks to create and maintain new types of services[10].Two important entities of an IN are the Service Switching Points(SSP s) and the Service Control Points(SCP s).The SSP s continuously receive requests of ser-vices which they cannot serve without the help of the SCP s where all service software resides.Thus,the SCP s are providers and the SSP s are customers.The SSP s and SCP s communicate via a signaling network which we here will represent as a cloud rather than a specific topology of signaling links and nodes.(See Figure1.)It is assumed that a small part of the available bandwidth of this network is reserved for the resource al-location,i.e.,the communication overhead caused by agent communication(and trans-portation).It is assumed that all SCP s support the same set of service classes and that all service requests can be directed by a SSP to any SCP.3.1Concrete multi-agent system architecturesWe have chosen one architecture of each of the abstract classes mentioned earlier(see table1).Common for these architectures are the use of three different types of agents: quantifiers,allocators,and distributors.A quantifier acts on behalf of a provider of the resources,an allocator acts on behalf of a customer,and a distributor decides the allo-cation of some(or even all)available resources.Although these three types of agentsSS7 Signalling Network "cloud". . . . . .SSP1SSPn SSP2. . . . . .SCPmSCP2SCP1Fig.1.A simplified view of an Intelligent Network (IN )withSCP s and SSP s (typically).centralizeddistributed synchronous Centralized auctions (CA )Hierarchical auctions (HA )asynchronous Centralized leaky bucket (CLB )Mobile brokers (MB )Table 1.The four different multi-agent system architectures classified in terms of distributedness and synchronicityhave similar roles in all the four multi-agent system architectures,the actual implemen-tation may be rather different (in particular this hold for the distributors).The reason,of course,is that different system architectures may put different demands on the agents.The centralized auction architecture The Centralized Auction (CA )architecture is an example of a synchronous,centralized architecture.Arvidsson et al.[1]suggested an approach where the resource allocation is carried out by means of tokens (cf.market-based control [5]).Each token represents a service request and is consumed when the request is accepted by a provider.The three types of agents have the following func-tionality:–The quantifiers try to sell the amount of tokens that corresponds to the load that theprovider is able serve between two auctions.–The allocators try to buy the amount of token corresponding to the resources itpredicts their customer will receive during the time to the next auction.–The distributor receives bids from the quantifiers corresponding the available ca-pacity at their provider (and the prices ),and bids from the allocators containing the expected need for resources.The distributor then carries out the auction so that the common good is maximized and sends messages about the result to the involved agents.An allocator maintains a pool of tokens for each provider and type of resource.Each time the allocator feeds a provider with a request for a particular type of resource,one token is removed from the associated pool.If all pools associated with a particular re-source type are empty,the customer cannot accept more requests.The pools are refilled at the auctions that take place atfixed time intervals.In order to avoid spending all tokens immediately during high loads(which would lead to excessive delays caused by long queues at the providers),percentage thinning is used so that the probability of buying a certain type of resource is never higher than the number of remaining tokens over the number of expected needs during the reminder of the interval.For more details we refer to Arvidsson et al.[1].The hierarchical auction-based architecture One possible implementation of a dis-tributed,synchronous system is the hierarchical auction(HA)architecture[11].The idea is to partition the set of allocators and to use one distributor for aggregating bids and holding auctions for each partition.These distributors then connect to higher order distributors in a hierarchical manner until the total demand can be matched against the amount of available resources offered by the quantifiers.The centralized leaky bucket architecture The centralized asynchronous architecture we have chosen is based upon an asynchronous approach called Leaky bucket[2].The basic idea is that each provider is equipped with a Leaky bucket that feeds requests to the provider at an even and optimal rate.This is done by inserting the incoming requests from the customers in a queue in the Leaky bucket.These requests are then dequeued at a rate corresponding to the maximum capacity of that provider.If the queue is full, the requests are rejected.To get a centralized architecture,we introduce a centralized leaky bucket(CLB)ar-chitecture,in which there is just one central distributor,common for all allocators and quantifiers.The allocators send all requests immediately to this distributor,which con-sists of a common leaky bucket for queuing the requests.It also has a router that contin-uously dequeues requests at a rate corresponding to the total capacity of the providers and then forwards the requests evenly to the providers in proportion to their capacity.If the bucket is full,the request is returned to the allocator where it is rejected.The mobile broker architecture As an example of a distributed,asynchronous system, we choose a mobile broker(MB)architecture[4].In this architecture,the distributors are implemented as mobile brokers(one for each provider)that sequentially visit each(or a subset)of the allocators offering the resources currently available at the corresponding provider The allocator then requests the resources it needs for the moment(or rather, predicts it will need in the near future).If possible,the broker gives this amount of resources to the allocator.Otherwise,it gives as much as is currently available at the provider.However,there are two problems with this naive approach:–If an allocator demands all the available resources,the broker will give them to that allocator.Thus,the broker will not be able to hand out any more resources for a while,which would not be fair.–If the overall load is low or moderate,the allocators are given just as much resources as they demand.However,if an allocator need slightly more resources than it asked for(predicted),it will have to turn down request,even though the provider has lots of surplus capacity.In order to solve these problems,we use a broker mechanism that strive to give out all the available resources and give each allocator resources in proportion to their part of the total current demand(of the allocators in the route).For the details of this approach we refer to Johansson et al.[7].It should be noted that in case of a sudden increase in demand,the resources given out may momentarily exceed the available resources, which in the worst case will lead to a transient overload situation.However,the mech-anism is self-stabilizing,and willfind an equilibrium within one route(given that the demands are relatively stable).The mechanism ensures that the allocators are given re-sources that(relatively)correspond to their share of the total demand handled by the broker,thus solving both problems in the naive solution above.If an allocator are visited by several brokers it may happen that some of the brokers’SCP are carrying a higher load than the others.To deal with this problem an additionalbalancing function is used,making the allocators try to move load from those SCP s with relatively high load to those with relatively low load.The allocator calculates the load of a broker from the quotient between what it asked for and what is was given by the broker.3.2Concrete attributesWe now operationalize the abstract attributes presented earlier.Thus,in the domain of IN load management the attributes are defined as follows:–Reactivity is measured by how fast the MAS is able to re-allocate the available SCP processing time when there are sudden changes of offered loads by the SSP s.–Load balancing is measured by the standard deviation between the carried load of the SCP s.–Fairness is measured by the standard deviation of rejected calls divided by the ac-cepted calls between the SSP s,i.e.,the rejection rate.–The utilization of resources is measured by how close the carried load is to the target load,or offered load,if the offered load is less than the target load.SCP load levels should be as close to the target load(e.g.,0.9Erlangs,corresponding to90% of its capacity)as possible but not exceed it.If an overload situation is approaching, the SSP s should throttle new requests.–Responsiveness is measured by the time it takes for the SSP s to get response from an SCP.–Communication overhead is measured by the bandwidth necessary for the MAS to perform the reallocation.–Robustness is measured in terms of the consequences of a distributor agent failure.–Modifiability is measured in terms of how easy it is to add a new or remove an existing SSP or SCP.–Scalability is measured in terms of how the number of SSP s and SCP s influence the performance-related attributes.3.3Experimental evaluationThe four concrete architectures have been evaluated in simulation experiments.For these experiments we used the same simulation model as Arvidsson et al.[1].The In-telligent Network modeled has8SCP s and32SSP s,which are assumed to be statisticallyidentical respectively.The processors at the SCP s are engineered to operate at90per-cent of the maximum capacity(target load is0.9Erlang).All messages passing throughthe network experience a constant delay of5ms,and it is assumed that no messages are lost.Detailed descriptions the simulation results can be found in[8].Since we usethese experiments as a case study,we only summarize the results here:–As the theoretical evaluation predicted,reactivity was promoted by the two asyn-chronous architectures.For instance,we simulated a scenario in which half of the32SSP s experience an offered load corresponding to0.2Erlang,and the other halfa load corresponding to1.4Erlang.The whole system thus is offered a total aver-age load of0.8Erlang,which is below the target load,and therefore possible forthe system to carry.At time400,the levels of loads are shifted from high to low and vice versa and10seconds later they are swapped back again,see Fig.2.The CLB manage this difficult situation perfectly and the other asynchronous architec-ture MB does almost as well.The auction-based architectures,on the other hand,have apparent difficulties to adapt to the changes.–All the architectures were able to balance the load well.There were just small differences in the standard deviations(of the measured carried load of the SCP s), varying between1and3mErlang depending on the offered load.When the offered load was below target load the centralized architectures performed slightly better, whereas there were no measurable differences in overload situations.–With respect to fairness,all architectures performed well when the offered loadwas below target load.In overload situations,MB was significantly less fair than the other architectures.However,it may be the case that an improvement of the design of the broker routes may reduce these differences.–It was not clear from theoretical analysis if there is any correlation between the abil-ity to utilize the resources and the architectural properties.The simulation results presented to the left in Fig.3indicates that centralized asynchronous architectures perform best in this respect.–Also,it was not clear the theoretical analysis if there is any correlation between re-sponsiveness and the architecture properties.But in this case the simulation results indicate that the centralized asynchronous architecture has the worst performance.See Fig.4.–We tuned the parameters in the simulation experiments so that both CA,HA and MB needed approximately the same bandwidth for communication overhead,about40 messages/second irrespective of the amount of offered load.The bandwidth needed by CLB,however,is proportional to the number of requests and is considerably higher than for the other architectures.For instance,when the offered load is0.70 Erlang,2150messages/second are sent,and when the offered load is2.0Erlang, 9365,i.e.,more than200times as much bandwidth as the other architectures need. For the general software architecture attributes(robustness,modifiability,and scalabil-ity),we refer to the theoretical analysis in section2.3.00.20.40.60.811.21.41.61.82400450O f f e r e d l o a d i n E r l a n g s Time in seconds SSPs 1-16SSPs 17-3200.20.40.60.81400A v e r a g e c a r r i e d l o a d i n E r l a n g s Time in seconds Centralized auction Distributed auction Mobile brokers Centralized leaky bucketFig.2.The expected offered load (left)and the average carried load (right).0.350.40.450.50.550.60.650.70.750.80.850.90.9510.40.60.81 1.2 1.4 1.6 1.82C a r r i e d l o a d i n E r l a n g s Offered load in ErlangsCentralized auction Hierarcical auction Mobile broker Centralized leaky bucket Min(Target load, offered load)Fig.3.The ability to carry offered when using the different architectures.00.020.040.060.080.10.120.140.160.180.20.40.60.81 1.2 1.4 1.6 1.82A v e r a g e t i m e t o c o n n e c t (s )Offered load in ErlangsCentralized auction Hierarcical auction Mobile broker Centralized leaky bucket Fig.4.The average response time for the connections when using the different architectures 4Conclusions and future workWe have described a systematic way of evaluating different aspects of different MAS architectures.This approach was applied to an abstract domain,namely dynamic re-source allocation,and a theoretical evaluation of abstract architectures was made.This was supplemented by an experimental study of an implementation of this abstract do-main,load balancing and overload control in Intelligent Networks.The experimental evaluation confirmed the conclusions of the theoretical analysis,e.g.,that asynchronous architectures are able to react faster than synchronous.In addition,it gave insights con-cerning attributes for which no clear conclusions could be achieved from the theoretical analysis,e.g.,that centralized asynchronous architectures are able to utilize the avail-able resources better than the other architectures,but have larger delays and need more bandwidth when the load is high.The results of the case study were,not very surprisingly,that different architectures excel in different dimensions.The choice of MAS architecture for a particular applica-tion should be guided by the balance of the trade-off between these dimensions that is optimal for that application.We believe that if the systematic approach suggested here is widely adopted,such choices can be more informed than is currently practice.Our plans for future work includes:–Further experimental validation in the IN domain of the theoretical results regard-ing,e.g.,scalability.–Experimental validation in another dynamic resource allocation domain.–Use the outlined approach to investigate other domains than dynamic resource al-location.。

相关文档
最新文档