The study explores Decentralized Uncoded Storage Elastic Computing (DUSEC) as an alternative to centralized systems, allowing any virtual machine to participate in computations. By proposing a decentralized storage assignment scheme, the study aims to optimize computation time and efficiency. The experiments conducted over the MNIST dataset using a Softmax regression model demonstrate the effectiveness of DUSEC compared to existing systems.
The paper discusses the importance of elasticity in modern cloud computing systems and introduces Coded Storage Elastic Computing (CSEC) as a solution. It highlights limitations of CSEC for certain types of computations due to linear coding requirements. The introduction of Centralized Uncoded Storage Elastic Computing (CUSEC) is discussed, focusing on heterogeneous computation speeds and direct data copying into virtual machines.
Furthermore, the study delves into the concept of Decentralized Uncoded Storage Elastic Computing (DUSEC), emphasizing coordination among different virtual machines' storage assignments. By proposing a computing scheme with closed-form optimal computation time under decentralized storage assignment, the paper aims to achieve state-of-the-art storage assignment efficiency.
The evaluation on Tencent Cloud platform compares DUSEC and CUSEC designs in terms of accuracy and computation time. The extension to straggler mitigation through encoding transmissions by each VM is also discussed, showcasing how elastic schemes can be combined with coding techniques for improved performance.
To Another Language
from source content
arxiv.org
Głębsze pytania