# 001 Buffering in Routers

# 001 Buffering in Routers

Many ASICs companies are developing ASICs to cater to different router needs. These router ASICs are either input-buffered, or output-buffered. Do we know what the main difference is and why we do it differently. The difference between input-buffered and output-buffered routers lies in where the buffering of packets occurs within the router. Buffering is used to manage congestion and ensure that packets can be handled efficiently as they move through the router. Here’s a detailed comparison of both type

?Input-Buffered Routers

In input-buffered routers, the packets are buffered at the input ports before they are processed by the switch fabric and forwarded to the output ports.

Key Characteristics:

  1. Buffer Location: Packets are stored in buffers at the input ports.
  2. Switch Fabric Access: Packets wait in the input buffers until the switch fabric grants them access to move to the output port.
  3. Head-of-Line (HOL) Blocking: A significant issue where a packet at the front of an input queue blocks the packets behind it, even if those packets could be sent to free output ports.

Advantages:

  • Simpler Output Port Design: Since buffering is handled at the input ports, the output ports can be simpler.
  • Potentially Lower Memory Requirement: Input buffering can be less memory-intensive in certain designs, as each input buffer only needs to handle the traffic from its own input port.
  • Less switch fabric bandwidth: As scheduling happened for input traffic, need to pass less traffic through the switch fabric and thus less fabric bandwidth requirement. Drop happens in input port itself.

Disadvantages:

  • HOL Blocking: Can significantly reduce the throughput and performance due to the HOL blocking issue.
  • Complex Scheduling: Requires sophisticated scheduling algorithms to manage the access to the switch fabric and reduce HOL blocking.

Output-Buffered Routers

In output-buffered routers, the packets are buffered at the output ports after they have been processed by the switch fabric.

Key Characteristics:

  1. Buffer Location: Packets are stored in buffers at the output ports.
  2. Switch Fabric Access: Packets are immediately transferred through the switch fabric to the output buffers if the output port has space.
  3. No HOL Blocking: Since packets are buffered after passing through the switch fabric, HOL blocking is not an issue.

Advantages:

  • Higher Throughput: Eliminates HOL blocking, allowing for potentially higher throughput and better performance.
  • Simpler Input Port Design: The input ports can be simpler since they do not handle buffering.

Disadvantages:

  • Higher Memory Requirement: Output buffering may require more memory because the output buffers must handle the aggregate traffic from all input ports.
  • Scalability Issues: As the number of input ports increases, the demand for output buffer memory also increases, which can be challenging to manage in large-scale systems.
  • More switch fabric bandwidth: As scheduling happened for output traffic, need to pass full traffic through the switch fabric and thus more fabric bandwidth requirement. Drop happens at output port.

?

The choice between input-buffered and output-buffered routers depends on the specific requirements of the network and the trade-offs that can be managed. Input-buffered routers can be simpler in terms of output port design but suffer from HOL blocking. Output-buffered routers can achieve higher throughput and avoid HOL blocking but require more memory at the output ports and can face scalability issues. Can routers be fabric-buffered and mix of all buffering, answer is yes. More to come soon on that topic…

#routers # buffering #HOL

Santanu Bhattacharyya

Design Eng Director at Xilinx || Datacenter SmartNIC || Networking || 5G || FPGA || ASIC

5 个月

Packet buffer sizing is another important parameter to consider. The trend across the industry, regardless of the application (switching/routing) demand, is to minimize buffering as much as possible without adversely affecting QoS. It relies on traffic engineering, advanced end-to-end congestion control algorithms, and active queue management (AQM) to notify congestion to the endpoints early and to keep the links saturated with small delay-bandwidth buffers.

要查看或添加评论,请登录

社区洞察

其他会员也浏览了