Distributed joint machine learning enables distributed computing power to jointly perform algorithm tasks, and deploys in large-scale distributed devices, effectively breaks through communication bottlenecks, and realizes joint computing of distributed computing power to complete data computing tasks. According to a white paper jointly released by IDC and Seagate, it is predicted that by 2025, the number of global connections will reach 100 billion, which is 10 times that of 2015. Global data storage has expanded 5.5 times from 32 ZB in 2018 to 175 ZB. Network computing power is ubiquitous. The exponential growth of global terminal node equipment makes the computing power scattered in different nodes of the network. The improvement of GPU computing performance makes each device a non-negligible computing resource, and is distributed in the global non-core. Large data centers in the region have an absolute cost advantage. The rapidly growing amount of data on the network is a resource pool with great potential. If you can combine distributed computing power, large-scale data and algorithms, it will create a valuable service market. Distributed AI provides the foundation from the algorithm level.
The blockchain provides economic dynamism for the cloud computing market in terms of data privacy protection and market incentives. In addition to user data privacy concerns, distributed cloud computing power creates an independent market bottleneck that also includes market incentives. Whether it is an individual or a company, it does not have the scale advantage of a large Internet company cloud computing platform. In essence, market incentives are the fundamental difficulty of distributed cloud computing resources not being put into the market on a large scale. The blockchain network solves the problem of data privacy. The blockchain account also provides natural market incentives for the cloud computing market, making the cloud computing market possible. In the cloud computing market, the blockchain ensures that user data is encrypted for the power resource provider. In essence, the blockchain acts as a decentralized system that supports the distributed cloud computing power supply and the market operations of user demand. Therefore, as a consensus mechanism for the core rules of market operation, its value and potential need to be further examined.
- An overview of encryption insurance: a huge business with an emerging valuation of billions of dollars
- Opinion: How much are you willing to pay for a better social network?
- Case editing | Which industries have been innovated in the blockchain of one night?
- DeFi Lending Overview | May 2019 News
- When the wind starts again, which is the central bank's digital currency?
- Many ministries such as the Central Bank and the Ministry of Science and Technology voiced that the blockchain continues to be supported by regulators
IDC in remote areas is expected to rise from the wind of cloud computing power, get rid of network restrictions, and take the cost of electricity and the ability to acquire customers as the core competitiveness. Under the continuous development trend of the cloud computing market, the network will no longer be a limitation, and the cost advantage of computing power will be highlighted because the data center power cost exceeds 50% of the operating cost. In the era of consumer Internet, the market pays attention to location advantages, management capabilities, and network bandwidth quality. In the era of computing power of industrial Internet, compared with the current mainstream Beishang Guangshen data center, the data center electricity cost advantage in the central and western regions is obvious. According to grassroots research, China The cost of electricity in the Midwest is reduced by more than 50% compared to the eastern region. For IDCs in remote locations, the ability to acquire customers will be key.
Investment suggestion: At present, the cloud computing market is still in the cultivation stage, and it is expected to explode in the post-5G cycle. IDC enterprises with high-quality cloud computing resources will be expected to share the cloud computing power market dividends. It is recommended to pay attention to companies such as Halo New Network, Meili Cloud, Data Port, NetScience Technology, Baoxin Software. At the same time, GPU and FPGA suppliers NVIDIA, Xilinx, Ziguang Guowei (002049) and Anlu Technology (unlisted) are also expected to benefit.
Risk warning: regulatory policy uncertainty, blockchain infrastructure development is not as expected.
We believe that the Internet will present a new business model driven by blockchain, AI and 5G convergence. The blockchain network provides a basic protocol for data market governance, assisting in the exchange of values between distributed network nodes. Based on the blockchain network, many data and resources have achieved distributed and market separation. Internet companies have reduced control over data, and algorithmic model providers have emerged. The 5G edge network will make cloud computing the next generation Internet infrastructure.
According to the thinking of the previous report, the blockchain provides unprecedented market incentives and economic dynamism for the Internet. We turn our perspectives into computing power, and explore what kind of market model and business value can be generated by the three important resource elements of the Internet, such as computing power, algorithms and data in the blockchain era.
From the perspective of computing power, based on blockchain, distributed cloud computing power and algorithms, data to achieve market separation becomes possible. Unlike traditional Internet giants that combine data, algorithms, and computing power, blockchains drive data, algorithms, and computing power to separate the market. For real-time sensitive business data, the traditional Internet giant solves the problem by central cloud + CDN service, but the potential of the Internet does not end with data transmission and access, and a large number of data computing and storage services are not sensitive to delay. This market is currently in a blank state. The evolution of AI algorithms and the rise of distributed AI will further drive more distributed intelligent services and newer business scenario models; further tapping the potential of data. In the past, large Internet companies represented by Amazon developed a new business model such as cloud computing based on their idle flexible computing resources. But this is still a kind of flexible reuse of centralized resources – Internet companies still master data, algorithms and computing power. In the mobile era, a large number of idle computing power and algorithm resources are distributed in many nodes of the global network. How to organize idle cloud computing resources to participate in market operations, so that idle resources generate market value – cloud computing power will become a new business model. The blockchain serves as the basis for data privacy protection, and more importantly provides the basic collaborative support for computing power and data separation. The rental of cloud computing power can obtain the required market return in the blockchain network.
A large number of IDC and personal computing node resources distributed around the world are not fully exploited due to network and algorithm bottlenecks. According to a white paper jointly released by IDC and Seagate, it is predicted that by 2025, the number of global connections will reach 100 billion, which is 10 times that of 2015. Global data storage has expanded 5.5 times from 32 ZB in 2018 to 175 ZB. With the exponential growth of data, very large data centers as the physical bearing of data become an important part of information infrastructure. At the same time, IDCs far from the network entrance and a large number of computing nodes distributed in the network become a new market protagonist. In the past, due to network delays and algorithmic efficiency bottlenecks, the Internet giant established a centralized data center with network export advantages to provide Internet services. Nowadays, in the business areas of data computing and storage, whether it is a large IDC or a more dispersed personal node computing power, there is a huge potential to be tapped.
Taking Google's joint learning algorithm as a typical example, distributed AI makes mobile distributed learning become a reality, and the potential of distributed computing power is activated. In February of this year, Google announced the realization of the world's first product-level ultra-large-scale mobile distributed machine learning system, which is now able to run on tens of millions of mobile phones. Based on TensorFlow, Google built the world's first product-level scalable large-scale mobile joint learning system, which is currently running on tens of millions of mobile phones. These phones can collaboratively learn a shared model, all training data is left on the device side, ensuring personal data security, mobile phone smart applications can also update faster and lower energy consumption. Researchers say the system is expected to run on billions of mobile phones. Joint learning produces a smarter model with lower latency and less power while ensuring user privacy. All of this relies on the Joint Learning (FL) approach, a distributed machine learning method that trains a large amount of scattered data stored on devices such as mobile phones, "introducing code into data rather than data." Introducing a more generalized implementation of the code and addressing basic issues such as privacy, ownership, and data location.
Distributed joint machine learning enables distributed computing power to jointly perform algorithm tasks, and deploys in large-scale distributed devices, effectively breaks through communication bottlenecks, and realizes joint computing of distributed computing power to complete data computing tasks. The exponentially growing global terminal node equipment makes the computing power scattered in different nodes of the network, and the improvement of GPU computing performance makes each device a power resource that cannot be ignored. The rapidly growing amount of data on the network is a resource pool with great potential. If you can combine distributed computing power, large-scale data and algorithms, it will create a valuable service market.
Network computing resources not only come from enterprises themselves, large cloud computing companies, but also the broader distributed cloud computing resources market needs to be further explored. The traditional Internet giant has mastered a large amount of data, algorithms and computing resources. The idle network computing resources are a waste, and the cloud computing service was born under the idea of flexible reuse. The investment of cloud computing services by traditional IT giants such as Amazon, Ali, Tencent and Huawei has made these companies pay a lot. But the more widely idle is the cloud computing power – after all, relying on the computing resources that the centralized Internet giants focus on is only part of the network computing power, we should not ignore the computing resources of a wider range of distributed nodes, these nodes may come from Individuals, small groups, or other companies that do not intend to set up a dedicated department to export cloud computing services. The model of large-scale cloud computing platform is still a centralized operation, and there are certain bottlenecks in business development and data privacy. Constructing a decentralized cloud computing market, which not only protects the privacy of the nodes but also the cloud computing power to obtain the desired market returns, is the potential for the cloud computing market to be tapped.
The Internet giant's cloud computing business is mainly based on consumer Internet companies, and more focused on real-time consumer-grade business. Cloud computing power is expected to become a new rising blue ocean market. From the perspective of the cloud computing market in 2018, it is mainly concentrated in the hands of Amazon, Microsoft, Google, Alibaba Cloud and IBM. According to Cannlys data, the above five giants account for 65% of the global market share, of which Amazon AWS accounts for 31.7% of the global market share. $25.4 billion. Ranked second is Microsoft Azure, with revenues of $13.5 billion, accounting for 16.8% of the global market. The head customers of Amazon's AWS business are mainly consumer Internet companies such as Apple, Adobe, Snap, Lyft and Pinterest. Other cloud computing giant customers are also concentrated in social and entertainment fields, and have high requirements for real-time interaction.
For large-scale data computing and storage services that are not sensitive to real-time network, cloud computing is expected to build a blue ocean market. Based on the Internet, it can be roughly divided into three categories: one is a real-time business, users need smooth access and browsing, such as web browsing, video entertainment, payment, etc.; one is a business that requires large-scale data computing. The business does not need too strong real-time, but it needs strong computing resources to support and carry out large-volume data calculation. Such services are represented by cryptocurrency mining, scientific computing, etc. Cold data services with low requirements for performance and computing power, such as storage backup. Among them, the latter two types of business scenarios are the most ideal market for cloud computing. This kind of business platform is currently in the early stage of development. Typical projects such as BOINC (full name Berkeley Open Infrastructure for NetworkComputing, Berkeley Open Network Computing Platform) were born in 2003. It is the world's oldest, most well-known and most user-distributed distributed computing. The internet. Currently, approximately 40 scientific projects around the world are using BOINC. The entire network contains more than 150,000 volunteers and 650,000 computers. These computers produce 30 "PetaFLOPS" computing power per second, or 3 billion floating-point arithmetic operations, comparable to the world's second-ranked supercomputer. If the Folding@Home project counts, the BOINC volunteers are more computationally intensive than any supercomputer. For large-scale data calculation, it is not sensitive to delay, and can fully mobilize the distributed cloud computing power to realize the cost-priority win-win situation of the resource provider's resource reuse and data calculation.
Encrypted digital currency mining has pioneered a rapidly growing market for cloud computing. Taking the mining of encrypted digital currency as an example, under the background of the bear market in the digital currency market, the computing power of Bitcoin's entire network is close to a record high. Behind these calculations are mines and mining machines distributed around the world. Most mines have extremely low requirements for IT environment and network environment. The main focus and cost of mining is electricity. Mobilize the global cloud computing power – whether it is a large IDC or a personal node, it can provide data privacy and efficient services for all kinds of data calculation. Encrypted digital currency mining has pioneered a fast-growing market for cloud computing, with certain demonstration effects. Leading the power consumption in mining to data due to business calculations, improving the consensus mechanism š algorithm, then equalizing the expansion of “mining” to a number of scenarios and areas outside the currency – the cloud computing market will be a bigger A more extensive mining market.
A number of blockchain projects have attempted to establish a cloud computing platform based on the blockchain network incentive system. Golem is a decentralized computer computing rental platform built on the Ethereum platform. Through the Golem platform, any user can become a power seller and renter. Whether you are providing an idle home computer or several large data centers, you can join the Golem platform. Golem is made up of all the nodes running the Golem application, running in a point-to-point manner, and can be used to simulate stock markets, big data analytics, medical research, and even cryptographic currency mining, which will restructure cloud mining businesses. The Golem program is to create a global, open source, decentralized supercomputer that anyone with access to the Internet can use.
In 2019, BOINC, the world's largest computing power platform, will open its new low-end blockchain transformation tour, which is expected to become the world's largest blockchain application project, and realize economic incentives through blockchain certification. Professor David Anderson, the BOINC project sponsor and scientist at the UC Berkeley Space Science Laboratory, plans to use blockchain technology to further advance BOINC. And convert BOINC into a P2P network with decentralized communication, computing and storage. In this case, the assignment calculation task can be done using smart contracts, and the scene space for recording all transaction information in the BOINC network using blocks is huge. At the same time, the blockchain-based Token incentive system will also provide a regulatory environment for the expansion of volunteer teams and computing power. Professor David Anderson believes that "in the concept of cryptocurrency in the blockchain, the Work Proof Mechanism (POW) is a useful calculation, not a meaningless hash function. There is already a Gridcoin integral incentive mechanism. We think this It's a great way to reward volunteers and attract new volunteers. This can increase the volunteer base by 100 or 1000 times, creating a new computing power that can revolutionize many areas of science."
The IDC in the marginal area is expected to rise from the wind of cloud computing power, get rid of the network restrictions, and take the cost of electricity and the ability to acquire customers as the core competitiveness. Under the constant development trend of the cloud computing market, the network will no longer become a limitation, and the cost advantage of computing power will be highlighted. In the future, low-latency services will be deployed in data centers in user-aggregated areas, while data backup, off-line analysis, and less demanding computing services will be preferentially placed in the climate-friendly, energy-rich Midwest data center. Compared with the current mainstream Beishang Guangshen data center, the data center electricity cost advantage in the central and western regions is obvious. Electricity bills are the core expenditures of data center operations. The cost of electricity bills accounts for more than 50% of total costs, which is a key factor in determining the success or failure of large-scale data center operations. According to the grassroots research, we compare the data center in the remote area with the current mainstream Beishang Guangshen data center. Take Lhasa as an example. The initial electricity price of the data center is 0.25 yuan/kWh, and the electricity consumption is gradually reduced. In Shanghai's large industrial power consumption, the electricity cost during peak hours exceeds 1 yuan/kWh, and the electricity cost during the valley period is also above 0.23 yuan/kWh. The cost advantage in the central and western regions is obvious, and the cost of electricity is reduced by more than 50% compared with the eastern region. For IDCs in remote areas, the ability to acquire customers will be key.
The PoW mechanism is the market power of the computing power market. The PoW mechanism is a proof of workload mechanism, that is, the competition for accounting rights (also the competition for the economic incentives of the general public) is to determine the winning and losing criteria through the competition of computing power. This is a typical free market mechanism, which is also the least wasteful situation from an economic perspective. In order to maintain the credibility and security of the network, it is necessary to supervise and punish evil nodes, prevent 51% attacks, etc., all under the constraints of the PoW consensus mechanism. From the most direct perspective, PoW is the most direct price mechanism in the computing power market.
In the era of cloud computing, the consensus mechanism is a market organization mechanism for large-scale Internet resources and value exchange, and its value needs to be re-examined. In the era of cloud computing, in the case of distributed deployment of Internet resources, large-scale peer-to-peer resources and value exchanges can be realized. At this time, it is impossible to rely on a centralized company to solve the huge and scattered supply and demand of resources. Blockchain is a market mechanism for large-scale peer-to-peer resources and value exchange on the basis of ensuring privacy. As a consensus mechanism for determining the core rules of this market operation, its function and value need to be further explored and examined.
From the perspective of the secondary market, according to our grassroots research, the current cloud computing market is still in the stage of demand cultivation, and most of the industry's Internet demand side is still in the self-built computing stage. We expect the cloud computing power demand in the post-5G cycle stage. It is expected to be gradually released. IDC enterprises with high-quality cloud computing resources will be expected to take the lead in sharing the cloud computing power market dividends. It is recommended to pay attention to companies such as Halo New Network, Meili Cloud, Data Port, NetSur Technology, Baoxin Software. At the same time, GPU, FPGA suppliers NVIDIA, Xilinx, Ziguang Guowei, Anlu Technology are also expected to benefit.
2. Blockchain infrastructure development is not up to expectations. The blockchain is the core technology for solving data privacy and economic driving in the cloud computing market. At present, the blockchain infrastructure cannot support high-performance network deployment. Decentralization and security will have some restraint on high performance. Blockchain foundation There is a risk that the facility will not develop as expected.
Source: Guosheng District Block Chain Research Institute