Windows Azure Handbook (6) Azure bandwidth with azure blob cloud storage

Windows Azure Platform Family of articles Catalog

　　In my recent years in the Azure presales work, I often encounter customers ask the same question: what is the bandwidth of azure virtual machines? Does Azure offer exclusive bandwidth? We need 200 trillion of exclusive bandwidth for this project.

When this happens, the author asks the customer: What is the purpose of the bandwidth you need to be exclusive?

Customers often answer: This app requires a video (large file) upload download function, or a large number of concurrent users, need to enjoy bandwidth to the corresponding more Internet requests.

This situation I understand very much, because we usually buy telecom broadband, is to buy 30M,100MB a year, the higher the bandwidth requirements, the more expensive price.

　　

For this exclusive bandwidth problem, the author wrote a detailed article, to introduce. The main content is divided into the following sections:

1. The drawbacks of exclusive bandwidth

2. Analyze Internet bandwidth Content

3. Related case sharing

　　

　　1. The drawbacks of exclusive bandwidth

In China, the cost of Internet access bandwidth is very expensive. I asked other colleagues that internet bandwidth in China is about 20 times times the cost of the United States. The price of exclusive bandwidth is obviously very expensive.

The most suitable scenarios for cloud computing include: switch mode, burst growth mode, and so on. buying exclusive bandwidth does not reflect the benefits of cloud computing resiliency. If you buy exclusive bandwidth, customers don't use cloud computing, and bandwidth costs must be paid. Assuming that the user purchased 200M bandwidth, the results of the project on-line to find a small number of users, the Internet bandwidth is idle, but this 200M bandwidth fees must be paid.

We assume that buying a cloud computing vendor's exclusive bandwidth of 200M can be seen from its official web site, which costs about $17717 for 1 months. Such as:

　　

While azure bandwidth is a shared bandwidth, I found in my project experience that the bandwidth of a single azure A3 VM (4CORE/7GB) is typically around 70MB.

3 Azure A3 VMs are load balanced on a vendor's bandwidth level, providing approximately 200MB of exclusive bandwidth levels. However, 3 Azure A3 VMs are priced at only $6071 per month. Such as:

　　

Microsoft Azure has a competitive advantage over the price of the cloud-computing vendor's exclusive bandwidth of 200M.

　　In addition, Microsoft Azure's bandwidth is increasing as the number of load balancer servers increases.

　　Add more azure VMs to your load balancer when the actual project goes live and discover that the interconnect bandwidth is not enough. When we find that the Internet bandwidth is idle, we shut down some of the azure VMs.

The benefits of this dynamic increase/decrease in azure virtual machines can help customers to significantly reduce costs.

　　

　　2. Analyze Internet bandwidth Content

　　Customers also said that our application needs to support 100,000 of users online video playback capabilities, we need exclusive bandwidth to ensure the user experience.

　　Let's analyze the scenario. When we browse a website, its content can be divided into the following parts:

(1) Static HTML page

(2) static files, such as videos, photos, documents, etc.

(3) Dynamically compiled pages, such as aspx,jsp, etc.

When a user accesses a video that is a host (Azure VM) bandwidth, there is no doubt that the larger the host bandwidth the better.

But please don't forget, Azure Block blob each file provides 60mb/s Internet bandwidth, and a storage account provides 10gb/s Internet bandwidth.

An Azure Block blob is similar to a cloud network disk.

We just need to change the software architecture:

-Dynamically compiled pages or go to host bandwidth

-static files, such as videos, saved to an azure Block Blob, such as the address: http://leizhangstorage.blob.core.chinacloudapi.cn/videos/1.mp4

-Static photo file, saved to Azure Block Blob, for example:http://leizhangstorage.blob.core.chinacloudapi.cn/images/1.jpg

Sending requests for dynamic content to an azure cloud host by sending a static content request to Azure Storage can significantly reduce the pressure on the cloud host to enjoy bandwidth alone.

　Next we say a case. It's me. A customer migrates the online training system to the Azure platform.

(1) Project background: Enterprise a online training system, mainly for employees in the enterprise online video training.

(2) Existing structure: the customer built the data center to purchase 60M exclusive bandwidth, all dynamic requests and static video files are going to the Internet bandwidth.

(3) Pain point: The 60M bandwidth cannot respond to large concurrent requests when the number of users who need training is exploding.

Customers also have security requirements for video files and cannot accelerate video access through CDN services. So you can only save video files to Azure Storage and set up SAS tokens with Azure Storage to control user access.

Before migrating this online training system to the Azure cloud platform, we also conducted an azure storage download stress test in 8 different cities across the country (Beijing, Shanghai, Guangzhou, Shenzhen, Chengdu, Hangzhou, Qingdao, Fuzhou), and the test results showed that azure Storage download speed is up to the maximum allowable download speed for your local network.

The organization diagram of the project migrating to the Azure cloud Platform is as follows:

　　

The project is a hybrid cloud approach that preserves existing input from the on-premises data center and saves video streams to azure storage in response to large concurrent requests.

The overall access process is as follows:

(1) An employee via mobile app, SSO single Sign-on to the IDC load balancer (load Balancer)

(2) The Web server that built the IDC data Center generates a token for the request and sends the token to the Web server deployed on Azure.

(3) The Azure Web server validates this token by returning it to the web on the IDC datacenter to prove that the request is valid.

(4) After token verification is passed, Azure Web accesses the storage account in Beijing and Shanghai, respectively, based on the business logic of the online training system, through the user's access ID.

If the user ID is from the north, the Azure storage generation SAS (Shared Access Signature) video URL in North China (Beijing) is returned to the client.

If the user ID is from the south, the Azure storage generated SAS video URL in East China (Shanghai) is returned to the client.

Finally, the video link address of the employee mobile app is actually the SAS URL generated by the storage in Azure Shanghai or Beijing.

Customer Benefits:

(1) The Azure Web server only verifies that the tokens sent by the IDC data center are valid. Therefore, video traffic does not go through the Azure Web server . Such as:

　　

You can see that in the last 7 days, the Azure Web server's output network traffic and input network traffic are not more than 75MB

(2) Azure storage is used to save the video file and return the SAS URL. Video traffic is passed through Azure Storage. Such as:

　　

As you can see, the total egress traffic for Azure storage in the last 7 days is 276.5GB.

Ticky:

In the previous content, I introduced the Azure Block blob each file provides 60mb/s Internet bandwidth, and a storage account provides 10gb/s Internet bandwidth.

What if the amount of user access is very large and exceeds the Internet bandwidth of a single file 60mb/s? It's simple, as long as we copy multiple copies of a video file.

We saved 6 copies of a video in the same storage account, and there was a total of 360mb/s Internet bandwidth.

I have 2 storage accounts in Beijing and Shanghai at the same time, the overall Internet bandwidth level is 720mb/s. This is amazing.

The author has developed a small program on the Azure Web server that distributes different requests evenly across different video files. Avoid all users accessing the same video file, resulting in bandwidth performance bottlenecks.

　　

======================================================== delimiter ==================================================== ======

Well, that's a good case. Let's talk about a less successful case.

An enterprise B migrates the platform to the Azure cloud platform as a whole, because its platform has 3 million fans, so the traffic is very large.

This involves the technical points of the load balancer, such as:

　　

Microsoft recommends a load-balancing mode such as the left side, by preventing multiple Azure VMs from responding to Internet requests on the back end of the load balancer.

Enterprise B's Software developers use only one azure VM as a front-facing machine, such as VM1. All user requests are sent to VM1, and then to VM2 and VM3 via VM1 's intranet IP.

There are 2 drawbacks to this architecture:

(1) Although the VM3 share the outbound traffic, but the VM1 of the public network inbound traffic will be very large.

(2) VM1 a single point of failure, if VM1 down, the entire application platform is not available.

In addition, the client underestimated the concurrent user's request and shut down the VM2 before the project went live. Only VM1 and VM3 are running throughout the architecture.

And the customer did not save the static file to azure storage, all requests are to take the host bandwidth, the pressure will be very big.

　　

The problem occurs immediately after the project is online. Such as:

　　

VM1 the input flow is 9.28GB and the output traffic bandwidth is 3.05GB in 5 minutes. The network is jammed directly into the dead. Service outage.

After this unsuccessful experience, it is summarized as follows:

(1) need to communicate with customers in advance, to do the online evaluation

(2) The disadvantage of using the host bandwidth of the file to describe the customer. It is also recommended that you save your static files to Azure Storage to avoid using host bandwidth.

　　

Windows Azure Handbook (6) Azure bandwidth and Azure blob cloud storage