云原生2.0加速:分布式云探寻全面云化的通路

{"type":"doc","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"企业数字化转型成为势不可挡的浪潮,尤其后疫情时代,各行各业都在积极数字化转型,谋求新的市场环境下的生存空间。据Gartner预测,2021年中国IT支出将达到3.09万亿,相比2020年增长7.7%。市场需求的不断升级,促使云基础设施服务市场快速响应。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"经历以“设备”为中心的服务器阶段,以“资源”为中心的云化阶段,企业数字化转型进程如今来到以“应用”为中心的云原生化阶段。降低人力依赖度,提高资源运维效率,降低部署繁杂度,提高应用便捷度,以及更加智能和安全可信,成为“云原生2.0”区别于此前的主要优势。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在4月8日的华为云TechWave全球技术峰会上,华为云对去年提出的“云原生2.0”进行进一步解读:全面云化和全栈智能是实现企业数字化治理的基石。换句话说,迁移平滑、运维稳定、开发便捷则是企业数字化转型的底色,分布式云x云原生在当下普遍被认为是企业实现数字化治理的有效通路。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"5月31日,华为云线上举办TechWave云基础设施专题日,基于云基础设施新范式进一步解读分布式云、云原生基础设施产品及背后黑科技,首次亮相分布式云操作系统瑶光。"}]},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"“一朵云”的关键特点"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"华为云分布式云具备四个关键特点:一张分布式网络满足便捷安全接入,一致运行环境构建无差别体验,一套管理体系应对精细化企业治理,一朵分布式云实现全场景覆盖。"}]},{"type":"image","attrs":{"src":"https:\/\/static001.infoq.cn\/resource\/image\/f3\/b4\/f3ff09336e086ab73a406fcc4b2c00b4.png","alt":null,"title":"","style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":"","fromPaste":false,"pastePass":false}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"具体而言,一张分布式网络,指的是实现企业IDC到公有云环境的便捷和安全接入,构建企业多种组网环境之间的高效互通;通过VPC云边内网互通,中心云服务与企业机房本地应用实现无缝互访,且可分钟级按需发放到边缘。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在华为云分布式云产品经理郑梁看来,这意味着物理分布,逻辑归一的分布式IT架构,典型场景如中心训练、边缘推理;本地数据脱敏后,还可以将数据回传中心数据湖,持续增加训练样本迭代优化。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这种分布式IT架构的实现,依托于华为云擎天架构,提供一致的运行环境。部署的便捷和运行的按需弹性服务,是分布式云给予的保证。在企业日常运营中,本地部署可以满足数据低延迟的诉求,一旦出现流量洪峰,本地数据库和中心Region协同工作,缓解流量压力。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"由于企业所处行业、组织架构不同,满足统一资源统筹和业务管理的诉求往往不易实现。华为云基于自身实践和对政企客户的深入了解,推出基于分布式云的管理体系,通过EPS和IAM的能力,构建一张整体管理视图,在人、物、财三方面实现精细化权限管理,匹配企业多层级、多部门的组织模型和管理诉求;通过RMS和CES能力,对整个分布式云统一资源管理和运维监控,进行精细化治理。 "}]},{"type":"image","attrs":{"src":"https:\/\/static001.infoq.cn\/resource\/image\/d6\/af\/d62cbf6f1dc8b395f6ccb92b694725af.png","alt":null,"title":"","style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":"","fromPaste":false,"pastePass":false}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在不同业务场景的部署上,这朵分布式云有多种表现形式:中心Region、智能边缘云IEC、智能边缘小站IES、智能边缘平台IEF和专属Region。如果按照网络的概念来类比,中心云类似骨干网的核心节点,智能边缘云类似于城域网节点,各类企业的现场边缘类似于接入网。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"其中,中心Region,服务丰富,满足集约效应和规模弹性;智能边缘云IEC提供小于10ms网络接入时延,实现热点区域的广覆盖和多线接入能力,满足就近接入及稳定低时延诉求,据郑梁介绍,IEC帮助国内某头部直播平台实现转码质量提高20%,带宽成本节省50%以上。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"区别于前者,智能边缘小站IES,把云服务延伸到企业数据中心,满足数据本地化和极低时延的要求,目前具备4节点起步的能力,可无缝拓展至16柜,支持10+云服务,提供小于5ms的本地时延,能实现整柜到货、4周内快速交付。中图云创通过本地部署智能边缘小站IES,在落地“全景新阅读”的沉浸式体验时,为VR设备提供端到端<30ms稳定低时延,节省30%的带宽成本。智能边缘云IEC和边缘小站IES的运维皆由华为云来负责。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"智能边缘平台IEF更轻量化,向上支撑应用,向下封装资源,目前支持包括电力、交通等行业在内的200+智能应用。在已经落地的全国高速取消省界项目中,实现了7.4万边缘节点,48万边缘应用“一张网”管理,比传统方式提效10倍,使车辆省界通行时间从15s降低到2s。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"专属Region主要面向大型政企治理诉求,提供全栈数字化转型底座。已经可以看到的实例是,中国一汽部署专属Region后,整体运维效率提升40%,通过使用全栈云服务,满足了企业办公、数字化营销、车联网、智慧出行等业务部署的要求。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这些服务的实现,底层扎根于华为云擎天架构,顶层则依靠华为云瑶光分布式云操作系统的统一调度、管理和分发。"}]},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"华为云的“智慧云脑”——瑶光分布式云操作系统"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"瑶光是华为云在2019年底发布的国内首个分布式云操作系统,相当于华为云的“智慧云脑”。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在整体设计上,华为云瑶光的关键架构包含多个层次:顶层是智能资源经营系统,二层是多目标优化系统,三层是全域调度系统,四层是对分式云站点管理,最小单元则是对多元硬件的管理,已经率先实现对x86、鲲鹏、升腾等多样性算力的兼容。这五层系统,基于端到端的仿真平台,进行算法的训练仿真与迭代,实现分布式云操作系统的自学习和自调优。"}]},{"type":"image","attrs":{"src":"https:\/\/static001.infoq.cn\/resource\/image\/47\/50\/47ddc6722cc8883536404f96a547e950.png","alt":null,"title":"","style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":"","fromPaste":false,"pastePass":false}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"瑶光的存在,目标实现全域资源供给极优,多样性算力使用极简。其中一大的难点在于,公有云租户对于云厂商们而言,动态变化的陌生租户业务类似一个黑盒,难以进行有效画像和预测。针对这个问题,瑶光构建了端到端的资源控制系统。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"其通过采集CPU微架构、利用率、操作系统内核等指标建立QoS量化模型,实时监测每个实例QoS干扰水平,然后通过将“对抗学习”和“知识迁移”相结合。其中,华为云自研的特征识别CDA-WGAN 算法能将业务特征识别准确度提升 10%以上,瑶光使用的自研METRO多维时序预测算法,与常用的开源算法相比能提升40%的准确,降低70%的误差。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"此外,为了真正实现“一朵云”的全域调度并保障确定性SLA,华为云开发了调度算法DQA。DQA 通过对网络服务质量、计算资源成本、云站点资源均衡度分别建模,并在收益函数中进行统一计算,帮助业务时延降低40%,同时成本优化达到33%。 同时,为了匹配动态业务诉求,华为云瑶光建立了包含机型设计、容量规划、资源分配等全链路仿真平台,为了匹配动态业务诉求,华为云瑶光建立了包含机型设计、容量规划、资源分配等全链路仿真平台。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"华为云瑶光将“复杂”留给自身,将“简单”交给客户。"}]},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"“简单”背后的黑科技,容器技术再革新 "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"长期以来,云原生落地的两大拦路虎是技术的和管理的复杂性,到目前为止的相关实践都是对这两大难题的不断解码。容器技术作为近两年的热门,是云原生基础设施的核心。为了实现计算、网络、调度上的全面进阶,华为云研发了容器集群CCE Turbo,并在今年为其装配容器卸载技术。 "}]},{"type":"image","attrs":{"src":"https:\/\/static001.infoq.cn\/resource\/image\/03\/ae\/03915f942070b6ba87457d24fbdaffae.png","alt":null,"title":"","style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":"","fromPaste":false,"pastePass":false}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"目前,华为云主要在3方面对容器引擎进行了优化,集群整体性能提升了40%:瘦身容器引擎,剔除冗余部件,优化内部执行逻辑;大量优化Kubernets中的Pod,如基于rust语言重写shimv2和agent,减少进程数量;大量优化容器引擎代码减少内存消耗,提升容器的启动性能和Cgroup管理能力。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在网络加速方面,CCE Turbo实现了容器到VPC的直接连通,弹性网卡Trunkport能力,并在此基础上对容器级的网络安全组的安全和效率进一步提升,新增安全组相关的Kubernetes CRD对象,简化容器安全访问规则配置。此外,华为云通过实现离线和在线业务的混合部署,及调度能力升级,提升集群资源利用率。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"值得一提的是,在调度方面,CCE Turbo实现了面向应用的智能化调度,通过IO Aware等调度算法优化、队列优化、Alloc_set等策略,提升大数据和AI作业调度效率,将AI、大数据类任务计算时间缩短30%~40%。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"此外,华为云此前联合合作伙伴共同发布云原生服务中心OSC,建立云原生服务生命周期治理规范并对接华为云应用市场。据悉,OSC规范下的云原生应用,可以发布到所有支持Operator、HELM标准的仓库中,同时,OSC还自动对接华为云容器引擎、多云容器平台、监控、告警等服务,实现应用跨公有云、专属云全球化无差异部署。目前,已经有130+数据库、中间件及行业应用基于OSC适配。 "}]},{"type":"heading","attrs":{"align":null,"level":1},"content":[{"type":"text","text":"基础能力全面升级,推新场景化解决方案"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"基于擎天架构,华为云基础能力迎来全面升级,在计算层面,发布全新的通用计算增强型云服务器C7;在存储层面,对SFS Turbo、Data+都进行了升级;在网络层面,发布独享型ELB和CLOUD WAN解决方案。其中,C7在视频转码、图像渲染等场景性能提升了30%,在MySQL等数据库场景性能上提升了50%,在Nginx等网络服务器场景性能上提升了60%。"}]},{"type":"image","attrs":{"src":"https:\/\/static001.infoq.cn\/resource\/image\/8b\/39\/8beb899b3f7eb79b52c7c3f6c1ce0639.png","alt":null,"title":"","style":[{"key":"width","value":"75%"},{"key":"bordertype","value":"none"}],"href":"","fromPaste":false,"pastePass":false}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"此外,华为云还推出两款全新场景化解决方案:异构计算三维建模联合解决方案和4K影视云制作解决方案。其中,异构计算三维建模联合解决方案由华为云和瞰景科技联合打造,建模性能提升30%,单次测量即可达到精度要求;4K影视云制作解决方案可以帮助节目制作的端到端效率提升30%,成本大幅降低,将大幅加速4K节目的普及。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":" "}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"帮助企业实现全面云化、拥有全栈智能的能力充满挑战,也是当下所有云服务商共同面临的长期课题。不局限交付位置,能够将公有云服务分发到业务所需位置,通过对不同节点统一管理,满足低延迟、本地化、多种混合场景的用户需求,分布式云x云原生的优越性已经显而易见。可以预见,分布式云x云原生的进展将在未来几年决定云的走向。"}]}]}
發表評論
所有評論
還沒有人評論,想成為第一個評論的人麼? 請在上方評論欄輸入並且點擊發布.
相關文章