Hello and welcome back to the course. Let's now take a look at CDNs architecture. Generally, when a client accesses the CDN, the CDN first determines which CDN node is closest to the client, and then guides the request of the client to the local server. The decisions are all made by the scheduling system. The scheduling system allocates requests of users to different nodes, based on multiple factors, such as location, ISP, traffic and port, and sets the routes of the requests in the CDN.

The traffic scheduling system and the optimal route selection system are core systems of Alibaba Cloud's CDN Service. The scheduling system detects network performative of links of all neighboring nodes every five seconds, monitors information about links to each node in real time, and guarantees each customer request can always be connected via an optimal route. Although it is possible to customize the schedule, we generally only do this if there's an extremely strong need or a high usage volume requirement that requires tweaking the CDN. Otherwise, we stick with the default settings here, and our content policies are already tuned based on our own strong experience operating a CDN.

So unless you, as a customer, have an extremely unique requirement, likely the default policies will be okay. Here on the slide, you can see there are level one cache clusters and level two cache clusters that form each CDN node. Every user request will first be routed to a level one cache cluster. And if there's a miss there, if there's not a cache hit, then the L2 cache cluster will try to provide the content. If the L2 cache also misses, a request will be sent to the origin instead. Here we show one type of CDN cluster.

There are three major different types of components here. There's LVS, Linux Virtual Server servers, there's Tengine servers, and there's Swift-based HTTP cache servers. So LVS performs layer four server load balancing. Why do we use the name Linux Virtual Server here? It's because clients interact with the cluster system as if it were a single, high-performance virtual machine, and this piece of load balancing software was written in Linux, hence the name, Linux Virtual Server.

As an advanced load balancing solution, LVS can be used to build highly scalable and highly available network services in Alibaba CDN, or using LVS direct route mode, which means all the replies will go back to users directly. We also use dual LVS to provide active, high availability. The two servers use keepalived, as you can see near the top of the diagram, to check each other's health status. Weighted round robin is used for server load balancing.

Tengine performs layer seven load balancing. Tengine is Alibaba Cloud's high performance HTTP server, which is developed on Nginx. It is an open source project which you can find on GitHub. Tengine will do proactive health checks, and provides SPDY, Speedy, Version 3 support. The real cached content will be stored on the backend servers which are running Swift. Tengine uses consistent hashing to try to balance the load to the Swift servers. The Swift servers would actually perform the HTTP caching. These servers use SSD desks to provide the optimal I/O performance. In the next section, we'll take a look at some of CDNs features.