Exclusive interview with Huang Huipan, CTO of Youpai Cloud: Advantages of technologically innovative CDN services

In the past two years, people have had a deeper understanding of CDN (Content Distribution Network). CDN technology has undoubtedly become a necessary support for the Internet. With the explosion of big data, driven by traditional Internet sub-industries such as video, e-commerce, and online games, and more and more traditional industries have deployed the Internet, China's professional CDN industry has shown a rapid growth trend in recent years. Therefore, 51CTO specially conducted this interview with Huang Huipan, CTO of Youpai Cloud.

[[127644]]

Huang Huipan, CTO of Youpai Cloud, author of aLiLua Web development framework. He joined Youpai in 2010 and started the R&D of Youpai Cloud storage related technologies. He has 12 years of experience in the Internet industry, and his technical experience covers a wide range. In the early days, he mainly engaged in front-end Web development (PHP/HTML/CSS/JS), and later gradually turned to the bottom-level R&D direction (Linux/C/Lua). He currently leads a team responsible for the R&D of CDN systems and bottom-level distributed storage systems.

When Huang Huipan first joined Youpai, he was mainly responsible for the CDN system. He combined the existing storage system to create a product, which became a USB cloud storage. Up to now, he is mainly responsible for managing the entire Youpai Cloud product department. Youpai Cloud has three products, one is CBI, one is cloud storage, and the other is cloud processing. Huang Huipan is currently in charge of all three lines. In the future, Youpai Cloud will focus on CDN, followed by cloud storage and cloud processing. CDN attaches great importance to the process of hardware investment and accumulation of hardware resources. Youpai Cloud has accumulated many years to achieve today's results.

Supply of entrepreneurial enterprise demand

Generally speaking, startup teams need a packaged product, which includes CDN storage and processing. Because the entire product of the startup team is basically started from scratch, the use of packaged products will have a significant acceleration effect. There is no need to develop many things, such as image thumbnails, audio compression, video processing, etc. Storage issues do not need to be considered. The most important thing is network acceleration. With the CDN of Yahoo Cloud, it is already a complete solution.

There are a lot of small files in the storage area, and many people are thinking about solving this problem. The performance of a large number of small files is very poor, and the poor performance is due to the IO problem, but the key is how you request these files and how to access them. If you provide access directly from the data center, the changes will be very large, and the storage cluster cannot handle it at this time. Unless those expensive solutions such as killing disks or killing card disks are replaced with SSDs, the IO problem of small files can be solved. Another way is to build a CDN network. What is the CDN network equivalent to? Youpai Cloud now has more than 100 nodes, each node has 20 to 30 servers, which is equivalent to hundreds or thousands of servers to carry these IOs, and these concurrencies will not reach a bottleneck that breaks the disk. However, if all these requests fall to the data center, it is also difficult to bear. At present, no one in the industry provides storage access services directly to the outside world, because this will cause IO problems. CDN is used, which is actually equivalent to a cache acceleration. In the 15-year plan, Youpai Cloud will build ten to twenty CDNs abroad, and Youpai Cloud's CDN network will have one of the best coverage rates in China.

Technical features of CDN system

In terms of technology, Huang Huipan said that the CDN system of Youpai Cloud is mainly written in C and Lua. The core group of the R&D team has ten people, 80% of whom are engaged in CDN, and Youpai Cloud's cloud storage is written (13:56). The cloud storage team has five or six people, and there is also a cloud processing line, which also has five or six people, mainly using Python and C. Many people don't know about Lua. It is the fastest among scripting languages, and no other scripting language can compare with it. Youpai Cloud began to study whether Lua is suitable in 2009. After more than a year of research, it was not until 2011 that Lua was really put into Youpai Cloud's business system. Youpai Cloud has certain technical strength and dares to innovate. The services and product functions it provides are all first proposed, so the technologies used internally are basically the first to try them out.

Why do CDNs choose these distributed computing service frameworks? The reason is simple: the development workload is small, the development cycle is short, and functions that meet business needs can be created quickly.

The CDN developed by Youpai Cloud is a little different from the traditional approach. Now it is truly based on the concept of cloud. The server at a specific node is actually just a grain of sand in the entire CDN network, a very small unit, and can be removed at any time.

It will not affect the entire CDN network. The traditional way of doing CDN is to divide the nodes according to the customers, but there is no such concept in the new generation CDN system of Yupai Cloud. All customers can access the same server and can do cache acceleration. Yupai Cloud will allocate thousands of servers to these customers for simultaneous use. For example, if customer A comes, you can use one thousand servers. If customer B comes, you can also use the same one thousand servers. If you randomly kill one hundred out of the one thousand, it will not cause any impact on the service, and it will not affect any customer.

In terms of system architecture, it is mainly divided into three layers. One is the edge acceleration node, one is the transit node, and one is the data center. Youpai Cloud has made a lot of optimizations to the CDN in terms of acceleration, with the focus on the transit nodes. Transit can avoid the problem of excessive pressure on the source station caused by many return-to-source processes. Because there are too many edge nodes, Youpai Cloud now has 100 nodes, and the next is 200 nodes. If these 200 nodes need to return to the source, if they return to the source individually, it will be 200 times of return to the source, but Youpai Cloud will have a transit agenda. There are ten transit nodes. These 200 return to these ten nodes, and then return to the source from these ten nodes, which is equivalent to only ten times of return to the source from 200. The technical characteristics of the transit layer are, first, multiple lines, and second, data, which is equivalent to a secondary data center. Its storage capacity is very large, already at the PP level, and the cache is at the PP level. The data center actually has 20 to 30 PPs now, and it caches one PP outside. Then, E-Pai Cloud has ten such transfer nodes, which is equivalent to caching 20 to 30 PPs, caching 100% of the data center data outside. Regarding cache distribution, it depends on the operation method. E-Pai Cloud uses the cloud method, with only one system facing all. The order of magnitude it generates is fixed. Even if customer A has a large website, it may only have a few hundred TB, which is impossible to reach the PP level.

Data migration in the CDN system is basically zero cost. You only need to do a very simple domain name resolution to migrate to UP CDN. As for cloud storage, there are some difficulties. The migration cycle will be very long, because the files are copied to the cloud storage of Youpai Cloud through the public network. The migration cycle varies according to the amount of data. It may be one month or one year. Youpai Cloud has also launched a mirror service to automatically store the other party's data so as not to affect the other party's service business during the migration process. There is another one that is more direct. We now have many customers who are like this. Because he did not use cloud storage before, he used his own server, and the content was all on the hard disk. He directly gave us the hard disk, and we went to the computer room to insert the hard disk. We will have a process of importing it into the cloud storage. Because we are in the intranet, the bandwidth is very sufficient, so its speed is very fast. If you have one or two TPs, it will be completed in two days.

There may be many difficulties when we first started to build the CDN system a few years ago. After several years of development, Youpai Cloud has basically overcome the technical difficulties and is now in a stage of mass deployment. When the system is mature and stable, the current focus is on the ongoing work of spreading nodes and then making the CDN network more perfect. CDN originally focuses on data distribution and does not require synchronization. However, when it comes to data centers, it involves synchronization, such as synchronization in computer rooms, which is easy to solve. The difficulty lies in cross-regional synchronization, because the stability and speed of cross-regional networks are too different from those of intranets. Youpai Cloud mainly has a dedicated synchronization system, which can schedule bandwidth according to actual conditions. During the busy business period during the day, Youpai Cloud will only use 20% of the bandwidth for synchronization, but at night when the bandwidth is idle, it will use 80% of the bandwidth for synchronization. This problem is actually not very technically difficult. It mainly depends on the bandwidth. It depends on whether the line is good. Youpai Cloud's current data centers basically have their own rented direct-connected optical groups. Therefore, Youpai Cloud has invested a lot in hardware, especially doubling the number of nodes in 2013 and 2014, and the goal is to triple it in 2015. At present, it mainly relies on some hardware to increase the coverage of hardware, because for Youpai Cloud's current network and CDN services, the redundancy is only 20-30%. If it wants to continue to expand the CDN market in the future, it must increase the redundancy of these bandwidths.

Innovative billing methods

Youpai Cloud has a unified billing method, charging according to actual usage, rather than purchasing a fixed amount and then charging every month. This will avoid waste. Youpai Cloud is also the first CDN service provider to propose such a concept, that is, charging by traffic, unlike the traditional method of charging by bandwidth. For example, if you purchase a 100-megabyte bandwidth and charge it according to the loan, it is possible that one day you will not do activities and only need 2 megabytes of bandwidth, but you will be charged for 100 megabytes. To be fair, charge by traffic. In fact, there are also customers with very stable businesses who require billing by bandwidth.

The domestic CDN market has huge potential and fierce competition

In the domestic market, professional CDN service providers have improved their competitiveness through financing and mergers and acquisitions to prepare for the increasingly fierce competition in the market. The CDN market will soon usher in a new development situation, and existing CDN service providers do not have an absolute competitive advantage. Although the technical threshold is relatively high, due to the fact that there is still a certain gap in talent, technical reserves, financial strength and market experience compared with leading foreign companies, domestic CDN companies need to invest more energy and make preparations in all aspects if they want to win in the competition.

Huang Huipan emphasized that Youpai Cloud is a company with CDN as its main product, not a cloud storage company. Youpai Cloud is very innovative in terms of technology and decision-making, and will not always stick to traditional methods. At the same time, he revealed that by the end of 2015, Youpai Cloud will have a relatively big move, which will be a relatively big subversive move in the Internet field, which will be different from the traditional perception.