Download PDFOpen PDF in browser

Peer-to-Peer Data Transfer Evaluation in SmartSSD-based Multi-devices System

9 pagesPublished: January 24, 2024

Abstract

Due to the growing demands on big data, data centers are expanding their systems to improve their capacity and capability. Common expansion techniques are: adding more memories, increasing the storage capacity, and attaching hardware accelerator devices. Benefit aside, such expansion puts a higher demand on the host system due to the increasing amount of data movements. It quickly consumes the available system bandwidth and sets a heavy burden on data transfers among the devices. To ease this situation, recent advanced solutions have appeared to optimize the data flow. These include peer-to-peer data transfer, allowing direct device-to-device data exchange without involving host memory. This paper evaluates the system performance for peer-to-peer (P2P) data transfer among connected devices. The results show that P2P data transfer between two devices in the system is 2x - 6x faster than non-P2P cases via bypassing the host memory. Not only does it reduce the memory occupancy but also the system power cost.

Keyphrases: data transfer, Multi-cores System, P2P, peer-to-peer, SmartSSD

In: Krishna Kambhampaty, Gongzhu Hu and Indranil Roy (editors). Proceedings of 36th International Conference on Computer Applications in Industry and Engineering, vol 97, pages 72--80

Links:
BibTeX entry
@inproceedings{CAINE2023:Peer_to_Peer_Data_Transfer_Evaluation,
  author    = {Luka Daoud and Gongjin Sun and Hingkwan Huen},
  title     = {Peer-to-Peer Data Transfer Evaluation in SmartSSD-based Multi-devices System},
  booktitle = {Proceedings of 36th International Conference on Computer Applications in Industry and Engineering},
  editor    = {Krishna Kambhampaty and Gongzhu Hu and Indranil Roy},
  series    = {EPiC Series in Computing},
  volume    = {97},
  pages     = {72--80},
  year      = {2024},
  publisher = {EasyChair},
  bibsource = {EasyChair, https://easychair.org},
  issn      = {2398-7340},
  url       = {https://easychair.org/publications/paper/Rnw1},
  doi       = {10.29007/lr5z}}
Download PDFOpen PDF in browser