Loading...

Delay Analysis and Improving in Distributed Computing

Doudi, Fatemeh | 2022

102 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 55757 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Ashtiani Mofrad Tehrani, Farid; Maddah Ali, Mohammad Ali
  7. Abstract:
  8. In this contemporary era, distributed computing is a vital solution for many applications. The traditional computing scheme based on allocating one complete computation to only one server is not efficient anymore. Delay has a very important role in computation’s performance especially for the computation at the edge of the network. To alleviate the effect of big data and complex algorithms, distributed computing is a very promising solution. In this structure, all computations are divided into some small tasks or coded, based on the system's status and will be executed in parallel. In fully parallelized distributed computing, a task is broken into N smaller tasks and whenever all N subtasks finish their job, the full computation can be retrieved. However, In full redundancy distributed computing, the whole task will be copied in all N servers and once the first subtask is received, the computation is completed. There are some cases, which use both parallelization and redundancy. These cases can be shown with (N,K) which means tasks are divided and coded and allocated to n servers and as soon as K subtasks complete their job, the overall computation can be recovered. In this research, our aim is to model the first 2 cases where we have full parallelization and full redundancy. We developed a model that can accurately predict the network delay using queueing networks. We evaluate the delay improvement of distributed computing compared to random allocating. We also showed the accuracy of the proposed analytical model by comparing it with simulation.
  9. Keywords:
  10. Distributed Computing ; Coding ; Queueing Network ; Queueing System ; Delay Computing ; Fully Redundancy Distributed Computing ; Fully Parallelized Distributed Computing

 Digital Object List

 Bookmark

No TOC