Automated deployment practices based on AWS

1. Background

In the past few years, social, mobile and cloud computing has profoundly changed the pattern of the entire Internet. As a global leader in the field of software design, Autodesk also formally began to transform from a traditional desktop design software provider to an online service, collaboration and mobile end design in 2009. In this transition, the company took full advantage of the huge advantages of modern cloud computing to bring customers much more than the traditional desktop software processing capabilities, user experience and cost-effective. In which AWS is currently the main operating platform for the company's services, the annual cost of tens of millions of dollars.

1.1. Challenges of traditional software delivery

Over the past more than 30 years, Autodesk has had a lot of desktop design software (such as Autocad,maya,3dsmax, etc.). Because design software often needs to deal with large data sets (such as the entire Shanghai Center design model) and extremely complex operational logic (such as Avatar Movie screen Drawing), its software size is generally large (GB level). In the past, the basic customer is through the Internet download, courier or distributors to obtain software installation package, the process is time-consuming. In addition, software upgrades, installations, and maintenance are also a huge workload (some large design companies are buying thousands of copies of software). Although the company software has already supported the centralized management model based on application virtualization, it still has obvious disadvantages such as the high cost of infrastructure construction, the lack of elasticity of service capability, and the need of the full-time personnel for operation and maintenance. So, improving the user experience of software delivery becomes a problem that we have to face.

As most people speculate, SaaS is our first attempt at the direction. In 2006, Autodesk Labs began experimenting with the possibility of SaaS design software services and achieved good results. However, currently SaaS applications still face the limitations of browser capabilities, large data transfer slow and many other problems, can not give professional designers the same experience and design of traditional desktop software.

1.2. New possibilities from cloud computing

With the advent of cloud computing, the infrastructure cloud provider represented by AWS has made it possible to solve the problem in a completely new way. We can take advantage of the powerful background of the infrastructure cloud to help users run virtualized software instances. Users do not need to download, install and maintain these software, just link to the Internet can use our software online. and pay on demand to reduce the cost of use. Based on this, the Autodesk Lab started this attempt in 2009 (note: The service has become part of the company's cloud platform product in 2013) and has chosen AWS as our background cloud service provider. There are several reasons to choose AWS:

Required Infrastructure Cloud (IaaS). Most of today's platform Cloud (PaaS) is prepared for Web services and is not suitable for us to run a virtualized instance.

Requires a strong elasticity of computing power. Autodesk design software for computing needs are very large, and the cost of computing power is not low. So, flexible computing ability can let us very good control cost. AWS EC2 In this respect is very much in line with our needs.

Requires a wealth of services. In addition to computing power, we also need to provide users with a large number of design data storage, fast data access, secure access control and so on. The AWS Cloud Service is very complete and easy to integrate with each other in this convenient service.

Need a stable service. AWS EC2 can provide more than 99.5% flexible computing availability, providing a solid infrastructure for building reliable services.

Requires global deployment. Autodesk is a company that provides software and services around the world, so we want infrastructure providers to be able to make a global layout. And AWS has built multiple data centers around the world.

2. Automated Deployment

After the basic implementation of the service and online services, the entire background maintenance and deployment costs are increasing. Building an automated deployment system is a must, especially given that we need high frequency (once every month), multiple (AWS) deployments of services and a high service availability.

2.1. Set goals

As a service running on AWS, we started thinking about the new possibilities and challenges that AWS brings to automated deployments at the beginning of our design. In our view, automating deployment of services on AWS requires special attention to some of the following features:

The service of the infrastructure. In AWS, you can use a similar cloud formation service to get all the infrastructure (including computing, storage, networking, IO, etc.) you want in a short time (a few minutes), and these infrastructures have been automatically configured and connected to your requirements. So we can get automated deployments to include infrastructure management, which is hard to implement in traditional datacenter mode.

Supports resilient resources. Because of the need to support resiliency, the entire service is dynamically adding new infrastructure, such as computing cells, at runtime. Automated deployment systems need to be able to put new infrastructure into service immediately.

Protecting data is more important. In the traditional data center, we can make the deployment complete in the internal network, after confirming all the security configuration and then online. But AWS is a public cloud service, and all of your deployments are actually done on a public network. Therefore, we must fully consider the issue of data security in an automated deployment.