3D-TTN: a power efficient cost effective high performance hierarchical interconnection network for next generation...

Area of Science:

• Computer Engineering
• Green Computing
• Network Architecture

Background:

• Green computing is crucial for reducing the environmental impact and operational costs of computer systems.
• High power consumption in supercomputers, particularly in interconnection networks, presents a significant challenge.
• Conventional interconnection networks often exhibit poor cost-performance ratios.

Purpose of the Study:

• To estimate and analyze the on-chip power usage and static network performance of 3D-Toroidal Triangular Networks (3D-TTN).
• To compare the power efficiency and performance metrics of 3D-TTN against other network topologies, including 5D-Torus, 3D-TESH, 3D-Mesh, 2D-Mesh, and 2D-Torus.
• To evaluate the cost and time efficiency of 3D-TTN through static effectiveness analysis.

Main Methods:

• Estimation of on-chip router power usage for different network topologies.
• Analysis of static network performance metrics such as diameter and average distance.
• Power-performance analysis, evaluating metrics per watt.
• Static effectiveness analysis for cost and time efficiency.

Main Results:

• 3D-TTN demonstrates approximately 32.48% less router power usage and improved diameter (21%) and average distance (12%) compared to 5D-Torus.
• Compared to 3D-TESH, 3D-TTN shows slightly higher router power usage (14.43%) but significantly better diameter (23.21%) and average distance (26.3%).
• Power-performance analysis indicates 3D-TTN outperforms 3D-Mesh, 2D-Mesh, 2D-Torus, and 3D-TESH even at the lowest network level.

Conclusions:

• 3D-TTN offers a promising solution for reducing power consumption in supercomputers.
• The hierarchical network architecture of 3D-TTN provides superior power-performance and static network efficiency.
• 3D-TTN presents a cost-effective and time-efficient alternative for green computing applications.