MassGrid Whitepaper

Wednesday, June 12, 2019
Download document
Save for later
Add to list

Decentralized High Performance Computing Network Vision With the rise of machine learning, big data, rendering services and search engines, high performance computing and parallel applica- tions are becoming faster and more demanding. The goal of MassGrid is to become the world's largest distributed GPU high-performance cloud computing network. MassGrid intends to transform the meaningless POW hash computing to general parallel computing that could be used for practical purpose through our improved POW algorithm and redesigned blockchain network architecture. rs Storage and Streaming Othe Media 32% 15% Non-IT industry Data Big applications 35% Cloud 18% High-performance High-performance Computing Computing15% Computing Scientific g5% Ot Computin Network he rs Services 45% 8% Artificial Intelligence 32% Cloud Computing High-performance Computing Road Map 2017.10 2018.6 2019.3 2017.11 2018.1 GPU virtual ma- Win/Ubuntu Ver- Masternode net- sion of the wallet CPU Miner Mobile Wallet work Alpha version chine network Alpha version 2017.9 2017.11 2017.12 2018.1 2018.10 2019.6 Test the network Jump POW GPU Miner Official website Masternode net- GPU virtual ma- prototype Algorithm and Mining pool wrok launched chine mainnet Technical Features Jump POW Algorithm GPU Computing Virtualization Smart Task Distribution Jump POW Algorithm is Mass- GPU DOCKER Virtualization Tech- MassGrid's P2P-based exchange Grid's patented blockchain tech- nology virtualizes remote GPU system is based on smart contracts nology, it can resist quantum hardware resources to a client’s and runtimes, users can customize attack and ASIC hardware take- devices. GPU DOCKER virtualiza- the functions and pay rules for their over, make the network safer, tion enables users to freely sched- specifc requirement, making trading fairer and guarantee all network ule hardware resources that are more flexible and extendable. nodes are capable to run general physically distributed around the parallel algorithm. world. What is Jump POW Algorithm Most cryptocurrencies are based on a fixed hash algorithm, no matter how complicated the algorithm is and how much space complex- ity it has. In the long time there is always the risk that mining could be replaced by ASIC hardware and finally centralized. Jump hash constantly changes as blockchain's data changes, Jump hash selects one hash algorithm from an algorithm pool which contains dozens of algorithms according to certain rules, the combination of hash algorithm which will be used constantly changes at each block. MassGrid will eventually integrate 30 different Hash algorithms, to design an ASIC for MassGrid mining, all 30 hash algorithms must be implemented in the hardware, only 1/30 of the chip resource will be used during computing each time, the rest will be idle. Taken together, Jump Hash significantly increases the GPU's energy efficiency relative to ASIC, as GPU power consumption and price goes down, GPU-based Jump Hash POWs will outperform ASIC economically. By implementing Jump Hash in our network, we can force all POW nodes to use the GPU for computing. Snefru Tiger Spectral FSB SWIFFT Hash Luffa PMAC Bmw Sha256 Blake Fugue Bmw Cubehash SipHash Groestl Shavite Scrypt MD6 Skein Simd Shabal HMAC Jh Echo Whirlpool GOST Keccak Hamsi Haval POW Hash Algorithm Pool MassGrid's algorithm pool picks 30 hash algorithms out of hundreds, combines the SHA3 family with several dozens other top-se- cure algorithms. All algorithms have been widely used and tested in many application before. Jump Hash algorithm GPU and ASIC comparison The programmable hardware structure of each comput- Only 1/30 of chip resource will be used in the ASIC ing unit on GPU dynamically switches algorithm to be because of non- programmable hardware structure, the able to achieve 100% hardware resource efficiency remaining resource are idle POW Computing Process Select algorithms from the Use the selected algorithm pool based algorithem to do the Hash Hit and return on data computing input blockData,lastBlockData; List hashIDs = GetHashIDByDataFactor(blockData,lastBlockData); temp = blockData; for( id in hashIDs){ temp = jumpHash[id](temp); } return temp; In our blockchain, every time a new data block needs to be generate, the Jump hash algorithm first extracts the data feature of the block or the previous block, and then selects one or several hash algorithm combinations according to the features of the data. Then uses this algorithm or a combination of algorithms for hashing, returning the correct result if hit or otherwise continuing. POW Verification Process Select algorithms from the Use the selected algorithm pool based algorithem to do the Hash Test results and return on data computing input hashResult ,blockData,lastBlockData; List hashIDs = GetHashIDByDataFactor(blockData,lastBlockData); temp = blockData; for( id in hashIDs){ temp = jumpHash[id](temp); } return temp==hashResult; When verifying the legitimacy of a data block, the Jump hash algorithm first extracts the data feature of the block or the previous block, and then selects one or several hash algorithm combinations according to the features, uses the algorithm or combination of algorithms for hashing, and compare the hash results to see it is correct. GPU Computing Virtualization Hash is currently the only POW algorithm with great features such as adjustable computational complexity, easy to verify, no data dependence and low data transmission. However, compared to hash algorithms , general-purpose computing’s complexity , data dependence , and data size varies with requirements, also it is not easy to verify, constraining general computing to fit POW can cause problems with poor generality of the computational network. By allowing users to access GPU resources in our computing network and allowing them to use virtualized GPUs to perform tasks and be paid based on hardware performance and usage time, we have therefor changed general-purpose parallel POW computing to a matter of virtualized GPU time-shared leasing. The advantages of a distributed GPU hardware virtualization network are: a) Compatible with almost all GPU-based computing tasks. b) Easy to calculate payment based on hardware performance and leasing time. c) Scaling flexibly on demand. We hope to introduce a cross-platform and efficient GPU virtualization protocol that will agglomerate various types of GPU devices into computing resources. In the future, we hope to optimize the x86 architecture so that any GPU devices following the virtualization standard without Host could directly connect in the network. node POW node user node Physical GPU device POW node Physical GPU node node Physical GPU Physical GPU device key node Physical GPU POW node POW node user Physical GPU Physical GPU Physical GPU device device node node node node The First Phase of MassGrid The Second Phase of MassGrid MassGrid first implements a blockchain p2p network based In the second phase, MassGrid will deploy multiple key nodes on the version 1.0 Jump hash algorithm that forces all nodes in the network. POW miners will register their physical GPUs accessing the network to use the GPU or CPU as computing to the key node. Key nodes will maintain a list of GPU devices. resources. Through remote virtualization, the POW miners' physical GPUs will be mapped to users, user will use these Among these nodes, some have GPU devices that efficiently resources and pay according to hardware performance and run Jump hash algorithms that perform POW computing and leasing time provide cryptographic verification services for the entire network. POW Resource Trade Process POW miners, computing users, normal users, and Key nodes are connected to each other through a p2p network , they will auto- matically to do Jump Hash POW mining by default POW miners will also register their physical GPU Key node, and the Key node will maintain a list of GPU resources When users initiate a GPU lease request, the request is submitted to the Key node. The Key node freezes the user’s budget by smart contract, establishes a virtualized connection for miner’s devices and users, and the connected GPU device stops POW mining , and begins to run the user’s computing job The Key node finally issues the frozen assets to the physical GPU host provider based on actual usage. Physical GPU First, each Keynode maintains a large list of POW nodes that auto- POW node matically register itself with Key node when it accesses the network. Users initiate GPU usage requests, submitting budgets and require- Send request user Keynode ments to Keynode, waiting for Keynode evaluations; Requests initiated include: GPU, CPU, memory performance require- ments, estimated duration of use, etc.; POW node Physical GPU Key node verifies the legitimacy of the request, assesses the data bandwides/computation requirements and esponds; Physical GPU POW node Scheduling resources If the request is valid, Keynode looks for the optimal node for the from POWnode user in the POW network according to the needs and the location of user Keynode Scheduling resources the user. from POWnode Optimal node is assessed according to the connection latency POW node between nodes and users, bandwidth and hardware specifications such as assessment; Physical GPU Physical GPU establish POW node POW node stop POW mining, and the Key node will establish a con- connection nection for the computing user and the POW node; user Keynode User gain full access to this virtualized device; The Key node finally issues the frozen assets to the physical GPU establish host provider based on actual usage. connection POW node Physical GPU Core Team Member Frank Lee Huang Xiang Perry Lei Frank Lee is one of the original Bitcoin After co-founding and acting CTO of a Full stack engineer and currently the gurus and has been in the cryptocurren- sports game company, Huang is a highly CTO of a Xiao Mi ecosystem company, cy community since the beginning. skilled expert and experienced in soft- Perry Lei has over a decade’s worth of Credited as the first miner to design ware development & management. The server side experience. In addition, he BTC & LTC mining chip , his knowledge depth and knowledge of his skill-set is was also an early investor in cryptocur- and experience in the industry is excep- essential for intuitive system design and rency and mining as a hobby since tional. integration. 2012. Liu Rui Hao Maxime Alexandre Dupuis Anonymity Liu RuiHao is an ACM award winner and Majored in Economics in 2012, Max coming soon algorithm researcher. He has a special has been involved in cryptocurrencies expertise in high performance parallel since its infancy. And was an early computing and p2p network. Liu and his investor in several top exchange web- team have helped several mining com- site and mining pools. Based in China pany deployed dozens of huge mining since after 2012, he has also been center across the world. involved in international business development for many years. Foundation Member & consultant Huo Ju Guo Hong Cai Wang Dong Huo is a technology pioneer and a One of the most famous investor in Wang is CTO of a financial company , famous tech column writer in China, he’s Chinese crypto currency community, He he has 10 years of experience in devel- deep insight in technology attracted is an angel investor of ETH and multiple oping large scale stock exchange several million subscriber and half billion successful block chain technology proj- system. Wang is also the leader of a big page view to his personal blog. ects. Guo also funded several of the tech team with more than 300 engi- world’s biggest cryptocurrency neers. The system his team running exchange website. handles and process es tens of millions deals per day.