Ali exclusive Secret Flying from research Roselle platform operating system how to prop up 268.4 billion global pro! | Asked at the end of China's IT technology evolution ...

Author | Zong Zhigang Ali cloud senior technical experts

Zebian | Tu Min

Exhibition | CSDN (ID: CSDNnews)

2019 Dual 11 day total turnover of 268.4 billion, the transaction peak 544 000 pen / second, behind volume data processing capacity of 970PB single day. Alibaba Cloud 100% on the core system, the first public cloud means that support the world's largest online trading, Ali cloud operating system from the research of Flying (Apsara), to complete the challenge. And "Luo" as the Flying virtual network system, it carries a very large scale concurrency during the double 11, super stability and complex traffic scheduling.

Below is a simplified electronic business system architecture, that is the bottom Ali cloud computing, storage, networking and other products and solutions. Since the electronic business platform requires enormous processing power, which relies on communications between distributed nodes, Roselle is a virtual network to provide.

Figure 1 simplified system architecture electricity supplier

What is Flying "Roselle" cloud networking platform?

Roselle Ali cloud flying distributed operating system is the core component that provides virtualized network services such as virtual private network (VPC), the software-defined network (SDN) controller and server load balancing (SLB) and many other networking products and technology.

How Roselle network platform supporting dual 11?

1, against Tbps-class traffic peak

At 00:02 on November the 11th, only began less than 2 minutes in the double-11, overall traffic peaked 1.1TB! Such a large concurrent access, is to buy buy buy 'chop hand party who "did not notice any difference, which is inseparable from behind Ali cloud network traffic peak two Titans: load balancing and SLB NAT gateway.

Load Balancing SLB (Server Load Balancer) is carrying Tbps-level traffic scheduling "housekeeper" completely "eliminate" slow site response or transaction failure phenomenon. By deploying a cluster, the average distribution of user requests and server by automatically removing aberrant health checks, eliminating single points of failure, improve system reliability. While achieving disaster recovery across the available area when the main area fails, the SLB second stage can achieve smooth switching operations.

当用户选好心仪的宝贝，跳转到支付宝进行付款时，就是由NAT网关调用支付宝的支付接口来实现，每一笔支付都会在NAT网关产生新连接。双11期间每秒交易峰值达到54.4万笔，就是背靠NAT网关超大规模的带宽和并发能力，其中单NAT网关的最大活跃连接数就高达400万。

2. 构建全球购物狂欢的高质量网络

2019年双11吸引了超过200个国家的全球用户参与，为了打造全球用户优质的购物体验，阿里巴巴集团在阿里云的河北，上海，深圳，香港，新加坡，美国等多个地域部署了Virtual Private Cloud(VPC)，并使用了弹性公网IP (Elastic IPAddress)，使不同地区和国家的用户能够就近、快速接入到网络。同时通过云企业网（Cloud Enterprise Network）进行多地域网络互联，分钟级构建出一张覆盖全球的，高质量的，安全的企业网络。

例如，新加坡的用户在购物时，就近接入新加坡阿里云并通过云企业网的专有链路到中国内地的阿里云，进而和部署在中国内地的阿里巴巴核心服务器和数据库进行交互。在这种跨地域的“专用高速公路”架构下，新加坡的用户可以获得极好的应用访问体验。

搭建好了全球网络，还需要超强算例资源部署。通过阿里云VPC ,工程师可以在1个小时内部署包括上万台云服务器。这在传统数据中心需要花费近月的时间。2019年双11云上单个VPC的IP规模达到 30万，最大的一个VPC的ECS和容器实例数量达到百万级，成为双11超级算力和最强大脑，云上公网和跨域出口流量达到数十Tbps，在业界遥遥领先。

3. 云网络QoS能力

双11期间，不仅是阿里巴巴电商购物，还有蚂蚁支付，菜鸟物流，大数据分析等多个业务系统均几乎同时进入流量高峰，然而不同系统对于通信的带宽、时延、丢包要求却不尽相同。为此，洛神网络平台提供了智能QoS功能，对不同业务进行流量优先级区分，对大带宽、丢包敏感业务设置通信报文高优先级，成为业界第一个具备QoS能力的云网络。

4. 智能网络分析

在双11的几个月前，洛神的网络运维和运营中枢—“齐天”已先行计算出承载双11流量所需的网络资源，精准的预测每个区域可承受的流量峰值、每条线路的带宽峰值、服务器数量。不仅如此，齐天还基于阿里云实时计算(Alibaba Cloud Realtime Compute)，阿里云时间序列数据库TSDB(TimeSeries Database)结合数据可视化，构建了一张全网实时监控大盘和分布式网络质量探测系统，支撑超过每秒2000万，多达100种核心网络数据的计算。通过智能诊断网络故障、多维度实时监测网络状态，齐天保障了双11网络的稳定运行。

2019 年双11，洛神网络平台强力支撑阿里集团核心业务的全面上云，提供快速、安全、稳定的网络服务。未来在弹性、开放、智能网络能力上会持续创新，提供给用户更加美好的体验，最终将做到让网络更简单，这是阿里云网络追求的理念也是使命。

系列阅读：