Scribd
Upload
9 views3 pages

NoC Research Synopsis

Uploaded by

Bhargavi V
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
9 views3 pages

NoC Research Synopsis

Uploaded by

Bhargavi V
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 3

Detailed Research Synopsis: "An

Effective On-Chip Network Topology for


Network on Chip (NoC) Trade-Offs"
Authors and Publication
Authors: M. Venkateswara Rao, T. V. Rama Krishna, S. Raaga Sai Sruthi, S. Akhila, Y. Gopi,
and L. Bhavani Krishna
Published in: Indian Journal of Science and Technology, Vol 9(17), May 2016
DOI: 10.17485/ijst/2016/v9i17/93120

1. Introduction and Background


The shift from traditional bus architectures to Network on Chip (NoC) architectures has
revolutionized communication within System on Chips (SoC). Traditional bus-based designs
are often limited in scalability and can suffer from congestion and bottlenecks. NoC, in
contrast, replaces buses with networks of routers and links that allow efficient packet-based
communication between interconnected components.

This study aims to evaluate four NoC topologies—Mesh, Torus, Dragonfly, and Flattened
Butterfly—by focusing on trade-offs between performance metrics such as latency,
throughput, hop count, and scalability.

2. Objectives
The primary objective of this research is to compare the performance of various NoC
topologies and identify the most effective one for efficient on-chip communication.
Specifically, the study addresses:
- Which topology offers the lowest latency and highest throughput?
- How do different routing algorithms impact performance under varied traffic conditions?
- Which topology can achieve optimal trade-offs between latency, hop count, and
throughput?

3. Methodology
The research uses software simulations to model and analyze the performance of the
following topologies:
1. Dragonfly
2. Flattened Butterfly
3. Torus
4. Mesh

The simulations are run with varying network radix (k) values (k = 2, 4, 8) to explore how
the performance of each topology scales with increasing network size. Key metrics used in
the performance analysis include:
- Latency: The time a packet takes from input to output through the network.
T = Th + Ts
- Throughput: The amount of data (in bits) transferred per second.
Throughput_ideal = b / q
- Hop Count: The number of routers a data packet passes through from source to
destination.

Graphs were plotted to visually compare the packet latency, hop count, and throughput for
each topology under varying traffic conditions.

4. Topology Descriptions

4.1. Dragonfly Topology


Dragonfly consists of high-radix and low-radix networks connected via global and local
links. Routers are grouped together, and global links connect different groups.
Advantages: Offers excellent path diversity due to the two-tier network design.
Challenges: Requires long cables for high-radix configurations, increasing latency.
Observation: Dragonfly had the highest latency (192.89 cycles) and runtime (129.7 units).

4.2. Flattened Butterfly Topology


This is an enhanced version of the traditional butterfly topology that reduces the number of
links, saving cost.
Advantages: High scalability and lower cost.
Challenges: Can introduce non-linear behavior at high-radix configurations.
Observation: Latency was moderate (25 cycles), and runtime was 17.5 units.

4.3. Torus Topology


Torus is an extended version of the mesh topology with wrap-around connections.
Advantages: Efficient path diversity and low network latency, making it ideal for large-scale
systems.
Observation: Torus showed the best performance with low latency (31.67 cycles) and
runtime (5 units).

4.4. Mesh Topology


In a mesh network, routers are arranged in a grid-like structure with neighboring
connections.
Advantages: Simple routing algorithms and easy implementation.
Challenges: Latency increases significantly with larger networks.
Observation: Mesh showed higher latency (62.82 cycles) and moderate runtime (4 units).

5. Results and Key Findings


The main trade-offs and findings are summarized below:
- Torus exhibited the lowest latency (31.67 cycles).
- Dragonfly had the highest latency (192.89 cycles).
- Torus and Mesh had similar hop counts, but Torus was more efficient overall.

6. Conclusion
The study concludes that Torus topology offers the best trade-off between latency,
throughput, and scalability. Dragonfly and Flattened Butterfly provide path diversity and
scalability but are more complex. Mesh is simple but struggles with scalability.

7. Applications and Future Directions


NoC technology is critical for modern systems like routers, data centers, and
supercomputers. Future research could explore:
- Wireless NoCs to reduce cable complexity.
- Adaptive routing algorithms for optimized performance.
- Hybrid topologies combining benefits from multiple architectures.

8. Summary Table of Results


Topology Latency (k=8) Hop Count Throughput Runtime (units)
Change

Torus 31.67 cycles 2.75 Moderate 5

Mesh 62.82 cycles 6.08 Moderate 4

Dragonfly 192.89 cycles 0.0 Low 129.7

Flattened 25 cycles 2.75 Moderate 17.5


Butterfly

You might also like