Hello and welcome to this Alibaba Cloud Academy online certification course on cloud computing. We'll be starting with ECS basic concepts. Let's first take a look at the Alibaba Cloud portfolio. Alibaba Cloud offers a wide variety of services in big data and AI, databases, middleware, video processing, IOT, application services, security, even private cloud. But we'll mostly be focusing on IAAS, or infrastructure as a service, in this course, starting with Elastic Compute. Alibaba Cloud's elastic computing Service is called ECS.

The Elastic Compute Service offers an SLA of 99.975% for a single virtual machine, which is very, very high, 99.995% for multi zone virtual machines, and our Elastic Compute Service is deployed in all 22 of our worldwide regions. Our machines all use modern Intel Xeon CPU's, sometimes with CPU frequencies up to 3.2 or even 3.8 gigahertz, and you can initialize or release up to 100 of these virtual machines in as little as 10 minutes. We offer many different classes or families of ECS virtual machine based on your workload.

So we have general purpose instances, compute optimized, memory optimized, high clock speed, local SSD, big data, and even an SAP instance type specifically designed to run SAP S/4HANA workloads. This is enabled by our third generation virtualization technology which we built ourselves which we call X-dragon, which gives us almost zero cost on the virtualization layer, so these virtual machines run as though they were your very own physical machine.

Billing for virtual machines is very flexible. You can purchase the machines on a pay as you go basis paying per hour, or you can purchase an upfront subscription for a month, two months, a year or even three years. We also offer a variety of heterogeneous instance types that are designed for workloads that require special purpose hardware. The heterogeneous computing types include instances that offer a GPU or an FPGA card. We also have Bare Metal instances which have no virtualization layer. These are a great choice for running your own virtualization scheme like VMware or Microsoft's Hyper-V. And then we offer HPC, which is a high performance computing instance type which can be used to build supercomputing clusters.

The advantage of the HPC instance type is that it has built in RDMA support and very high-speed convergent ethernet connectivity. These instance types are designed for things like very high speed media transcoding, industry simulation, rendering, or even genetics. If you're still not 100% sure what ECS is, this next part is for you. ECS stands for Elastic Compute Service, and this is a flexible on-cloud computing service that you can use to launch virtual machines, that's what it does.

An ECS instance is essentially just a virtual machine that lives on the Alibaba's public cloud. The advantage of this is you get a lot more flexibility, stability and security than you would get in your own cloud environment because you benefit from all the work Alibaba has done to improve and grow its public cloud offerings, to improve its own security, and to provide stable service. So an ECS is a virtual machine that you run on Alibaba cloud. If you compare ECS with a traditional off-cloud deployment, the key benefit is operations and maintenance costs are lower.

So for an ECS instance, physical server management, storage, physical networking configuration, power, cooling, and virtualization, these things are all Alibab cloud's problem. You don't have to worry about them. You only worry about launching your virtual machine, logging into it and setting up the software that you want. So you get lower O&M burdens by moving to ECS. Another key advantage not mentioned here is of course flexibility. Unlike a physical server, you can create a VM and destroy a VM anytime you want without waiting.

If you buy physical hardware, you're then committed to that hardware even if you end up not using it. So the cloud deployments can in fact be cheaper than off-cloud deployments. What really makes ECS work is virtualization. So in Alibaba cloud we of course have multiple data centers around the world. In each data center, we have racks and racks of physical servers. Each physical server is running Linux, and then it's running a virtualization layer.

In many cases, it is Kernel virtual machine, or KVM. When you launch an elastic compute service instance, an ECS instance, what's happening is we are creating a KVM virtual machine and launching it on one of these physical servers inside of our data center. Because your workload is living within this virtual machine, it can be moved around between physical machines as needed, which helps us make sure that your compute service is fully reliable because if a physical server fails, we can simply migrate your virtual machine to a different physical server that's healthy.

In order to do all of this migration and management and decide where to schedule your virtual machines, we have to have a management layer running on top of our cluster of Linux machines inside of our data center. We call this layer Apsara. In Chinese it's called feitian. Apsara is our self-developed distributed system for managing our cloud platform. All of our products are built on top of Apsara. Essentially, Apsara handles storage, scheduling, resource management and distributed coordination for all of our products that run on our public cloud.

You can see some of the components of Apsara here in orange, and then our public cloud products running on top in gray. In particular, Pangu, our distributed file, is a key component of all of our storage services, including storage for our ECS instances. So when you allocate an ECS instance disc, that's actually running on top of a distributed file system, which gives you extra reliability over and above what you would get with a single physical block device.

Currently Alibaba cloud has 22 regions across the world at 66 zones. ECS is available in every single region and zone no matter where you are in the world. So US West, US East, our data center in London in the UK, in Frankfurt in Germany, in the Middle East in Dubai, in Hong Kong and our many data centers in China, all of them support ECS. So wherever you happen to be, you can make use of ECS. Within each of those regions, we have multiple data centers.

So for instance, in Hong Kong or in Singapore we have multiple zones. Each zone is a physical data center with independent power and network connectivity, fault isolation can easily be performed between zones so if a single zone fails, the other zones in the region are not effected. Because each region contains multiple zones, it's possible for you to build failure-tolerant architectures by deploying multiple ECS instances into each zone. For instance, in Singapore, if you're running a web application, you could deploy an ECS instance in Singapore zone A, and then a clone or a twin of that ECS instance in zone B. If zone A were to fail, the instance in zone B can continue to serve requests.

How do you decide whether or not to build a multi zone architecture? Well, you have to make a trade off between network latency and high availability. If you want high availability and you can tolerate a slight network latency, then you go with a multi zone deployment. If you have to have the absolute fastest communication possible between all of your VMs, then you put them all in a single zone. So it's important to understand the network topology, the connectivity between regions and zones.

So within a single region, all of the zones are connected together over Ali-Baba Cloud's private network and traffic that passes between the zones within a region is free. So if you communicate over our private network between an ECS instance in zone A and an ECS instance in zone B, both in Singapore, then all of that network traffic is free. However, if your instance in Singapore needs to talk to an ECS instance in Beijing, that will not be free. You will incur a public bandwidth charge for using the public internet or a VPN gateway charge for having a VPN tunnel or a CEN network charge if you choose to use our enterprise multi-region private network. We do have a private network between our regions, but we charge you for bandwidth if you use that multi-region private network. That's a major distinction.

Again, within a single region, zones can communicate with each other over the private network for free, between regions, you either have to use the public internet, a VPN gateway on top of the public internet or our cloud enterprise networks, CEN, which is a paid private network product that connects multiple regions. So when you're planning your architecture, plan ahead what region you should place things in. If you are going to build a multi-region deployment, think about how you're going to connect those regions together.