Wu Pengcheng: Rapidly build a global IPFS and blockchain platform

Wu Pengcheng, East China Manager of Amazon Cloud Solution Architect, gave a speech entitled "Quickly Build a Global IPFS and Blockchain Platform", pointing out that the combination of data storage and IPFS requires grasping three points, the core of which is the global layout capability, followed by massive storage capacity, and finally unlimited public network, bandwidth and IP addresses.

The following is the guest speech content:

Wu Pengcheng: In today's situation where railways are not yet fully popularized, if you want to go to a place today, it is best to drive yourself and take the highway. Railways make money by selling train tickets, and train tickets cannot be sold next to the railway, but must be sold next to the highway.

You need to set up a lot of stores. The first point we can talk about today is how to combine the public cloud with our IPFS under the current situation. Then we talked a lot about mining just now, and we can also talk about how to do it. The second point is that after the mining is completed today, to cash it out and turn it into legal currency or a product that can generate income, it requires an exchange. Let's talk about how to build an exchange, so we will have some brief sharing today.

The first point is to build an IPFS storage system on the cloud . In fact, when building the storage system, some of the points that we pay close attention to are what are the core points when we build the IPFS network?

The first very important point is the global layout capability. If we build this network today to serve all the networks in the world and allow everyone to freely store data in the IPFS storage network, the core point is that it must be a global layout. If we make mining machines today, it is also very important to reach all continents in the world.

The second point is massive storage capacity . The core purpose of what we are talking about today is storage. Various structured and unstructured data need to be stored in the cloud, which requires massive storage capacity, which can be obtained from the cloud whenever needed.

The third point is unlimited public network, bandwidth and IP addresses . In the field of bandwidth, whether it is mining or later data storage, it can be retrieved by others. Because you are retrieved by others, they have to pay you with tokens. These will require these storage systems to have strong enough public network bandwidth capabilities. They cannot only provide one or two hundred megabits of bandwidth like some public cloud vendors. I think this is very important.

Of course, it includes the public network IP. When launching a peer-to-peer network on the Internet, it is very important to recognize each other. Uninterrupted operation is indispensable. When mining, you need to use File to make a deposit in advance. If there is a terminal, it will also lead to a great loss if mining fails. This is also very important.

Another thing is that we also need strong computing power during the mining process. For example, there will be some calculations during the data initialization process. A relatively powerful CPU can help you complete the initialization. When you officially launch the storage system to provide external services, it is very important that GPO is no longer needed. How to flexibly mobilize GPO and CPU? Acceleration is only achieved during initialization. This is a natural embrace of the public cloud because it meets our five aspects.

This is a world map. We can see many white and blue dots on it. These are global layouts, which means that there will be corresponding dots in all continents, including Africa and South America. Japan, Jakarta, and Spain will gradually open some in the next one or two years. If it complies with the rules of Bitcoin today, the issuance of other currencies will also have a geographical logic. If the mining machines are deployed globally, the content of the global layout needs to be considered.

In addition to the computing nodes deployed around the world, it is also very important to connect them together. Amazon provides a global network technology that connects every point in the world. There are no two pink points in China because they comply with some Chinese regulatory requirements. The two points in Beijing and Ningxia are isolated from foreign countries. We will talk about connecting them later, and there will be some compliance with China's technical and regulatory requirements.

So when designing from the perspective of stability and reliability, we need to consider that the first cloud platform must have sufficient redundancy. If there is a problem in the entire data center, the light is cut off, or there is a power outage, we can ensure that our business system can be run in another batch of computer rooms. The intermediate time should be as small as possible. This may involve the construction of some underlying infrastructure. These are some of the work we do in paving the way and building bridges.

In addition, when the machine itself has only a hardware failure, we do not want employees to intervene to maintain the health of the machine, so we hope to make the machine self-recovering as much as possible and able to repair itself. The host hardware failures we listed above, as well as memory exhaustion, some operating system crashes, etc., all require automatic recovery.

Amazon Cloud will help you restore functions. These are some interfaces. From this perspective, when mining with a cloud host, the risk of having your tokens deducted is much smaller.

If we want to use cloud hosts for mining today, the key points are how to choose machines. This option includes the ones you need with more CPU, more memory, more storage, and some GPU models, etc. At the same time, you can plan and design the storage capacity according to your needs. It is completely flexible. We provide 270 types of machines to choose from.

I can recommend a few here. In Filecoin mining, computing power is very important. As you can see, AMD's CPU often has some advantages over Intel's at the underlying level. In many scenarios, we will use AMD's processors for mining, so you may see many mining machine manufacturers on the scene.

It is also because AMD has such capabilities. AMD is divided into three types, one is T3 . If you want to see how it runs, we can use this one. It is a shared type. If you don’t want to ensure that 100% of the machines are equipped with 4 CPUs, or as many as you need, you will use the M series and R series. The M series is general-purpose, and the R series has higher memory requirements.

If AMD is not enough to meet your performance needs, you can also use GPU-accelerated ones , such as G4 and P3, which we provide on the cloud. These can help you quickly initialize the block.

There is also Elasic . If you want to speed it up, just during the initialization phase, you can bind a GPU to it. After the initialization is completed, the GPU is not of much value, so we will remove it. This process only takes two clicks of the mouse on the cloud.

In addition, we can have ARM processors, which may be traditionally inferior to Intel or AMD, but now we seem to have developed to the eighth generation, and many have surpassed Intel and AMD. ARM has many application scenarios. For example, some of the models of our latest ARM processors can reach 25GB of bandwidth. This is actually very suitable in the Filecoin scenario.

This is a comparison of several models. You will see that the ARM processor model adopts a 7-nanometer process, which is actually the most advanced in the world. Each chip has 64 cores, which is also more than the previous one. There is a cache on the CPU, which is much larger than Intel's. ARM processors can replace Intel and AMD CPUs in many scenarios.

This is a simple performance comparison we made. In fact, you can see it directly from the picture. The blue one belongs to ARM's test results, and the gray one is Intel's. We can see that in some scenarios, the ARM processor performance of integer calculations is nearly 50% higher than Intel's. When the floating-point budget increases by more than 30%, the ARM processor will be very good.

In addition to computing, we also need to look at storage. In this storage area, we will provide various types of storage, such as object storage, block storage, file storage, backup storage, including a series of underground storage. If you like to use FTP or SFTP, you can use them. These are storage capabilities that can be obtained with just a few clicks of the mouse on the cloud. There is also a code library. The core of this code base is that I don’t need to use these block storages now.

Because it is relatively expensive, it is not possible to store it on S3, so IPFS officially provides such a plug-in. After the plug-in is installed on the front-end server, the data can be stored in S3. S3 can achieve TB-level throughput, which is completely impossible to do normally. There is no need to worry about data storage issues. This is replaced with the front-end computing node. The original data can be hung on another server and work normally. The cost is about 70% lower than standard block storage. If anyone here is interested in mining, you can consider it.

Just now we talked about how to build an exchange based on the top exchanges in the world, if the mine is out and dug out. We can also see what options are available when we are doing blockchain applications today. We say that blockchain is so-called decentralization, but the essential purpose of decentralization is that it cannot be deleted or modified and no one is trusted. Let's discuss how to avoid tampering. If there is a centralized mechanism today to make the database completely tamper-proof, technically completely tamper-proof, this can replace some of the needs of some distributed scenarios. The blockchain method exchanges time for space.

In fact, the performance efficiency is not very good. We have a so-called centralized classification database called QLDB, which can ensure performance efficiency while making data tamper-proof. There is also a decentralized platform that provides a managed blockchain chain based on the Kaiyuan protocol, which you can drive with just a few clicks of the mouse.

If you go to the exchange today and focus on the core content, you can build some mining pools, which is also related to the previous mining stage. There are several leading exchanges in China, and they are like this, including the technical application of blockchain itself, the application of data currency, and some related industry applications, transaction settlement, and some anti-counterfeiting contracts, etc. There are also digital currency trading platforms, as well as some high-frequency transactions, algorithmic transactions, spot and futures, which can also be linked here.

In fact, when we talk about exchanges, the core will need to start from the five aspects that meet Maslow's needs. Everyone has heard that a person has five major needs in living in the world. The first is survival, the second is security, the third is to be understood, the fourth is to be respected, and the fifth is to realize self-worth.

When you run an exchange, you will also have these five major needs. The first survival need is to realize basic business functions. The second security need is that exchanges involve a currency, such as Bitcoin, which is very valuable. As long as you are popular, people will be jealous, so many people will attack you and try to steal. There is also the need to be understood and to understand the needs of others. This actually requires a lot of big data, such as acquiring customers and retaining existing customers. These all require big data analysis. There is also the need to be respected. Don't think my platform is stupid, so there must be risk control.

The demand for value realization is to output technology. Generally speaking, exchanges have five aspects, and the core needs to solve five problems. Our core projects are like this. At the beginning, we have to build a platform for matching transactions, then do security protection, big data analysis, risk control, and finalize the product. After it is done well, we can do custody. There are many similar companies in the exchange that specialize in this field.

When building an exchange, we have some reference architectures. For example, starting with user access, we need to first improve and verify it, and finally ensure that it can resist external attacks. Exchanges are attacked on a daily basis. How do we prevent others from doing some attacks, such as malicious attacks at the grassroots level, and then to the back-end trading platform? The digital market fluctuates very strongly, and its peaks and troughs vary greatly. If at certain times, the market soars and there is a big difference in the market, we need some flexible processing capabilities.

Including back-end data storage , it will also involve some wallet applications. These will require you to lose some data, which means the loss of coins and money. There will be some core technologies to solve these problems. Including big data business , this will also use a series of analysis services to solve the so-called understanding and being understood needs just mentioned.

The issue of security has received a lot of attention in the cryptocurrency circle . Once coin theft occurs, which happened to many companies last year, the loss is very large. Some reasons are, for example, because others attack you, and some are caused by malicious scripts or viruses. In fact, this is done in five aspects: identity and access control, and detective control.

If you want to build some of your infrastructure in China, in this case there is a technical solution that says that the site you build overseas, whether it is a server or something else, some of the exposed IPs can do some routing acceleration, so that a certain effect can be achieved.

To sum up, as the largest service provider, Amazon can have very good cooperation with railway builders. For example, we can have very close cooperation in mining and exchanges. Mining can provide various GPUs, CPUs, and various types of storage to achieve low-cost and fast use scenarios. In the exchange scenario, the five major requirements mentioned above can be quickly implemented with a good architecture.

Thank you everyone. This is the end of my speech today. We have a booth outside where you just came in. If you are interested in learning more, you are welcome to go to the booth for detailed communication. Thank you everyone.

Wu Pengcheng, East China Manager of Amazon Cloud Solutions Architect, attended the 2020 IPFS Ecosystem and Distributed Storage Industry Summit on July 3 and delivered a keynote speech entitled "Quickly Building a Global IPFS and Blockchain Platform". The above is some selected content.

——END——