高级检索
    衷璐洁, 王目. 区块链赋能的算力网络协同资源调度方法[J]. 计算机研究与发展, 2023, 60(4): 750-762. DOI: 10.7544/issn1000-1239.202330002
    引用本文: 衷璐洁, 王目. 区块链赋能的算力网络协同资源调度方法[J]. 计算机研究与发展, 2023, 60(4): 750-762. DOI: 10.7544/issn1000-1239.202330002
    Zhong Lujie, Wang Mu. Blockchain-Enpowered Cooperative Resource Allocation Scheme for Computing First Network[J]. Journal of Computer Research and Development, 2023, 60(4): 750-762. DOI: 10.7544/issn1000-1239.202330002
    Citation: Zhong Lujie, Wang Mu. Blockchain-Enpowered Cooperative Resource Allocation Scheme for Computing First Network[J]. Journal of Computer Research and Development, 2023, 60(4): 750-762. DOI: 10.7544/issn1000-1239.202330002

    区块链赋能的算力网络协同资源调度方法

    Blockchain-Enpowered Cooperative Resource Allocation Scheme for Computing First Network

    • 摘要: 随着AI内容生成、多媒体处理、VR视频等对于计算资源有着极大需求的互联网服务的快速发展,在可以遇见的将来,计算资源将成为网络中的稀缺资源.算力网络通过将算力作为网络基本单元之一来实现算力的网络化,为这些计算敏感的应用提供了行之有效的解决方案.得益于来自云—边—端等节点的计算资源,算力网络能够为大规模用户提供弹性泛在的计算调度.尽管算力网络具有广泛的应用前景,如何实现在这些地理分布的计算节点之间高效调度,计算资源对于算力网络的性能至关重要.提出了一种区块链赋能的资源调度(blockchain empowered resource allocation,BCERA)算法.不同于现有的资源调度方法,BCERA依赖于一个区块链结构来实现分布式、高效的计算资源调度.特别地,有别于现有的区块链结构,BCERA中的区块链节点通过求解任务调度优化问题来实现区块链的共识,从而在避免共识所带来的额外开销和时延的同时,还能提升系统的可扩展性和鲁棒性.计算资源调度问题被建模为一个马尔可夫决策过程(Markov decision process,MDP)并通过强化学习方法来求解.除此之外,还设计了一个激励机制以鼓励计算节点贡献资源支持算力网络的运转.实验结果表明,所提出的方法能够有效提高资源调度性能.

       

      Abstract: With the fast rapid growth of the Internet services such as AI content generation, multimedia processing, VR vidio etc., with huge demand for computation resources, it is forseen that the computing will become scarce resources in the near future. Computation first network (CFN) conceptualizes networked computing and puts the computation as the primitive of the network, which becomes a promising solution for various computation intensive applications. CFN provides ubiquitous computation access thanks to the resources offered by various computation unit from the cloud, edge and end users’ equipments. Despite its promising, how to offload the continuous arrival computation tasks to various geo-distributed computation units in CFN is of importance to the CFN performance. In this paper, we propose a blockchain empowered resource allocation (BCERA) for CFN. In BCERA, blockchain plays a key role for the task offloading, including recording the resource usage, optimization of the resource allocation. Specifically, we redesign the consensus mechanism in blockchain and allow the blockchain nodes to reach agreement via solving the resource allocation problem in FCN. This resource allocation problem is formulated as a Markov decision process (MDP) and blockchain nodes use the reinforcement learning methods to search the optimum. Besides, considering the resources may be offered from different parties, an incentive mechanism is also presented to encourage the computation units to provide their resources. We conduct a series simulation tests based on our designed blockchain platform to show how BCERA outperforms the state-of-art solutions.

       

    /

    返回文章
    返回