Editor’s note: This article is from WeChat public account “Brain Polar Body” (ID: unity007), author Tibetan Fox, 36 氪 released with permission.It came, it came, it came with an AI solution!The overseas epidemic situation is becoming increasingly severe, and overseas versions of AI devoted to fighting the epidemic have also been launched.In many projects, shared computing power schemes sound particularly familiar.It’s a bit like the Chinese people “lay at home and contribute to the country”, Nvidia called on gamers to “open a computer to contribute to the epidemic.”, I urge PC players to donate their idle GPU / CPU computing power to support the distributed computing Folding @ home (FAH) project of Stanford University to make up for the lack of computing power in virus research.Giants and netizens such as Intel and MSI also responded, saying that they were ready: in order to save the world, pay more for electricity!MSI even used the “Avengers” final battle map, using “everyone is a superhero” to suggest crazy.However, can shared computing power help fight the new crown virus?Why is it not seen in China’s epidemic prevention program?Shared computing power: The nationwide version of distributed computing wants to understand the value of shared computing power against the new crown virus, starting with Folding @ home initiated by Stanford University’s Pande Lab.The so-called computing power sharing refers to relying on the established network information center and using cloud distributed computing technology to realize the interconnection and sharing of network computing power to achieve analysis and calculation.For example, Folding @ home, which is used to fight the new crown epidemic, is currently the world’s largest distributed computing project (Guinness World Records in 2007). It is mainly used for AI research on diseases. It is used for protein folding, aggregation and the resultingFor related diseases, perform computational drug design and other molecular dynamics research.Since its launch on October 1, 2000, it has attracted the participation of business giants such as Nvidia and Sony to successfully simulate the folding process of 5-10 microseconds.At present, all GPU projects at Folding @ home are also targeting COVID-19, aiming to find its potential drug target, and the CPU project will be added as soon as possible in the future.For the player’s participation method, you need to select “Any disease” on the project client, and you will receive related requirements, and you can set the client according to it.Then Folding @ home will be in the background with a very high priority, that is, to provide cases when the computer is idle, there is no need to worry about affecting the normal computer experience such as games and videos.And before the outbreak, there have been many mature applications of hashrate sharing.The project “BOINC Computing Power Earth”, established as early as 2002, claims to be able to help ordinary people’s computers achieve “the fifth type of contact” with alien civilizations.C-side users allow BOINC to call personal CPU and GPU computing power when idle, supporting scientific research in various fields such as mathematics, medicine, astronomy, meteorology.The University of Oxford in the United Kingdom has used BOINC’s computing power to predict global climate change in the next 100 years.The computing power of the BOINC platform has continued to grow over the past decade.As of March 2019, more than 4.4 million volunteer users have joined BOINC, and there are about 600,000 mainframes daily, contributing about 30 PFLOPS of computing power.If calculated on the basis of the same amount of computing power rent of AWS server, the value of sharing one year is equivalent to the global PC users donating 50 million US dollars for scientific research.The founder of BOINC is David Anderson, a well-known scientist in the field of distributed computing at the University of California, Berkeley.The largest project supported by BOINC is SETI @ HOME, a computing resource sharing program for searching for intelligent alien creatures, initiated by the University of California, Berkeley.Analyzing the radio signals collected at the Arecibo Observatory and Telescope in Puerto Rico and the Green Bank Telescope in Virginia to find evidence of the existence of extraterrestrial intelligent creatures is a huge one.Computing Engineering.Since its launch on May 17, 1999, it has attracted more than 5 million users worldwide, more than 710,000 active hosts, and provided about 30 PetaFLOPS of computing power per day.Of course, the distributed computing platform really enters people’s vision on a large scale, and it is inseparable from the market baptism of “mining.”From 2017 to 2018, the popularity of blockchain (especially digital currencies) has also driven the “shared computing power economy”, and many platforms have emerged that can lease personal computer resources.Users host the machines with graphics cards in the mining pool, and lease the remaining computing power to the “mine” for mining, and the platform can reduce the computing power cost of virtual coins.So, what are the advantages and disadvantages of shared computing power as a “democratized version” of distributed computing?Plump ideals and skinny reality: the true face of shared computing power The king said, “Dear Minister, I dreamed of a number last night, which is 190554261410902619. I don’t know if this number is a prime number. I need to know the answer as soon as possible.”The Minister replied: “Your Majesty, I don’t know, but we have just given each citizen in the kingdom an ID number in natural order. As long as we issue an order to let everyone use his number to remove the number that the king dreamed of,The answer will be obtained soon. “The next day after the order was issued, the king received two reports, one was 456275009 and the other was 456275291.This story contains the idea of distributed computing.On the one hand, the rapid development of information technology and the wave of industrialization of artificial intelligence have made computing power for processing and analyzing massive data a new key resource.For example, projects such as protein analysis involve complex model structures and huge calculations. Even the use of supercomputers takes a long time and high server rental costs.If many computers can be involved in the calculation process, the items that need a lot of calculations are divided into small pieces, which are processed by multiple computers at the same time, and then the results of the calculations are uploaded and combined to draw data conclusions, the process can be greatly shortened.Moreover, unlike a centralized commercial cloud computing platform, a large number of users of personal computers have some idle resources.Statistics show that the annual shipments of computers worldwide are 200 million units. Based on a five-year replacement cycle, there are about 1 billion computers worldwide running at any time, but the utilization rate is only 20-30%.Most of the time it is idle.If you make them nodes for distributed computing, and donate them in the form of public welfare donations or small-cost purchases, those projects that have been stalled due to inability to pay the computing power will be supported and used on demand. Wouldn’t human technological progress have to be accelerated a lot??Of course, sharing computing power has made its debut for many years and has never reaped large-scale individual user support. There must be a special reason behind it.First, shared computing power generally occurs on projects with high public welfare attributes or projects with high economic benefits.Individual idle computing power is indeed huge in resources and low in cost. Who would not want to “respond to the situation” in this computing power enclosure movement?After all, cloud service providers build their own data centers, and they also need to invest in the cost of machine rooms, rent, electricity, operation and maintenance.If you let the whole people work for yourself, you think about this scenario … you want to be beautiful!You know, even if the reputation of the platform is guaranteed, users must continue to run the computer when sharing idle computing power, especially when using GPU computing, it will always remain at full load and the power consumption will inevitably increase.Some will also occupy software memory and cause stalls, which will also reduce the durability and life of the device.Therefore, unless there are sufficient reasons to drive, such as fighting against the virus for the community of human destiny, or giving sufficient economic returns, such as mining, it is difficult to mobilize the enthusiasm of most people.Secondly, even if it is a worthy project / platform, it may not be able to manage the shared resources.On the one hand, distributed computing is only suitable for those studies that can solve all or part of the problem through computing, and the computing process needs to be easily segmented into acceptable sizes for personal computer processing power, which limits many studies to adopt the “shared model”Come on.On the other hand, the project party / platform party needs to manage the GPU / CPU hardware resources of a large number of users. However, due to commercial competition, the design details of the GPU have not been disclosed, and there are also large differences in products from different manufacturers.Moreover, uploading computing power to cloud virtualization also results in performance loss, which all increases the difficulty of its deployment and management.This is why, 21 years after going live, SETI @ Home decided to terminate assignments to volunteers on March 31 this year.The project team explained that because all the required data had been analyzed, and the distributed computing management of the data was laborious, the ad hoc group decided to focus on completing the back-end analysis of the data and writing the paper.Third, a platform that maximizes the efficiency of shared resources must have strong technical capabilities, which also limits the expansion of shared projects.Because of the cloud virtualization and deployment of massive personal computing power, a distributed computing environment (also called middleware) needs to be deployed to provide public services and support distributed applications. Otherwise, project staff will have to solve multiple operating systems,A variety of network protocols, a variety of databases, performance, efficiency, security, etc. have no direct relationship with the business itself.For example, the emergence of virtualization technology, multi-core CPUs and GPUs with a large number of cores, while increasing computer performance by orders of magnitude, has also increased the difficulty of shared deployment, which can easily cause performance confusion and waste of resources for GPU-intensive workloads.Virtualization is needed to abstract and simulate rich computing resources, so that the computing power can reach the performance of the native GPPU / CPU, while thousands of applications do not have any interference with each other.This powerful, stable and unified technology is mainly in the hands of cloud computing vendors.This may also explain from a certain level that in the AI anti-epidemic action in China, the option of “shared GPU for all people” has not yet appeared.Emerging, with its own responsibility: One of today’s situation in China’s cloud computing, China’s cloud computing industry continues to grow in scale and has relatively sufficient computing power resources.In the past few years, from national policies to corporate needs, it has driven the rapid growth of the domestic cloud computing industry.According to data released by research institutes such as the Institute of Information and Communication Technology and IDC, in 2018, the scale of China’s cloud computing industry reached 96.28 billion yuan, an increase of 39.2% over 2017. The industry scale in 2019 is expected to exceed 100 billion yuan, reaching 129.07 billion yuan.Many provinces and cities are building supercomputing centers, which provides a prerequisite for efficient strategic mobilization for the shortage of AI computing power during the epidemic.Second, the market share and technical capabilities of Chinese cloud service vendors are at the forefront of the world, and they have stepped forward to open up their computing power during the crisis.IDC’s “Global Public Cloud Service Market Tracking” report shows that the overall market size of China’s public cloud services (IaaS / PaaS / SaaS) exceeds USD 4 billion, and Chinese cloud vendors occupy the top four seats in the world.The joining of these technology companies also underpins the demand for computing power.Alibaba Cloud announces that all AI computing power is freely available to global public research institutions, Baidu Research Institute is free to open Linear Time Algorithm LinearFold and the world’s fastest RNA structure prediction website; Didi Cloud has also freely opened GPU cloud computing resources and technologiesSupport, for combating epidemic-related work … In addition, some cloud service vendors are also actively researching and developing, and freely open a number of intelligent service products specifically for epidemic investigation, investigation, prevention and control, such as at least Ali, Tencent, Byte Beat,Huawei and other cloud collaborative office vendors have successively opened many functions for free. The Shanghai Economic and Information Commission and various operators have negotiated to provide free services for cloud office and cloud video conferences for more than 6 months … There are so many computing powers that support computing.The hard-core input of resources naturally does not need to call on the whole people to start and support the epidemic.From this perspective, the “GPU fight against the epidemic”, which is a race against time, is not only a strong support for scientific researchers racing against the new coronavirus, but also the best profile of the global cloud computing industry..
Why did Nvidia’s “GPU Sharing Anti-epidemic Law” fail to enter China’s plan?
36 氪 starterAutopilotBlack technologycapital marketcareducationEntrepreneurshipForefrontFrontlinegameGo publicInformationInternet of ThingsinvestmentNet red economyNew businessNew economyOccupationreal estatestockTechnologyTelematicsthe Internetunmanned