Abstract:Emulating output queued (OQ) switch with sustainable implementation cost and low fixed delay is always preferable in high performance router design. Distributed shared memory (DSM) switch, also called switch-memory-switch (SMS) router, provides a promising way towards emulating OQ in backbone switches. However, the previous architectures and algorithms for DSM/SMS switches require either high implementation cost or iterations of scheduling. This paper proposes a Buffered Crossbar based DSM switch (BC-DSM), which can emulate the FCFS OQ based on a speedup-two buffered crossbar. Each crosspoint has only two packet buffers, one for arrival packet and one for departure packet Different with previous works of DSM, we do not aim to resolve memory access conflicts when packets arrive, but to randomly dispatch packets into memories through buffered crossbar and resolve departure conflicts by bounded small queueing and delay. For scheduling, the departure packets are prioritized in the transferring between the fabric and the memory. The advantages of our scheme include: 1) Only N two-speedup memories and one two-speedup buffered N times N crossbar are needed. 2) Scheduling process is simple and distributive, thanks to buffered crossbar. 3) No need for additional crosspoint buffer for overcoming long round-trip time (RTT), thanks to the probabilistic load-balancing scheme. 4) Simulation shows that only a few fixed delay is needed to provide extremely low packet loss rate such as 10-8, under different types of traffic patterns, including bursty model and unbalanced model.

Research of the High-Speed Fibre Switch Network NIC Based on the Memory Communication

Design of the DDR SDRAM Controller Integrated by the Data Intensive Computing Architecture

A 2-Gb/s/pin Source Synchronous CDMA Bus Interface with Simultaneous Multi-Chip Access and Reconfigurable I/O Capability

Architecture design and performance analysis of a novel memory system for high-bandwidth onboard switching fabric

Memory Centric Interconnection Mechanism for Message Passing in Parallel Systems

A high performance inter-chip fiber communication scheme with short frame protocol

Design of High-Speed Data Transmission System Based on Fiber Channel

High Performance Packet Processing with FlexNIC

Design method for fiber channel application system

Design and Implementation of an AHB-based High-Speed Memory Access Layer

Design of Dual-Shared Dram Controller Based on Switch

Performance Comparison of Cell-Based and Packet-Based Switching Schemes for Shared Memory Switches

Obtaining High Performance Switching with Port Distributed Memories

Understanding I/O Direct Cache Access Performance for End Host Networking

Analysis of Tri-Stage Memory Array for High-Speed Packet Buffers

SAMNoC: A Novel Optical Network-on-Chip for Energy-Efficient Memory Access

Emulating FCFS OQ with Buffered Crossbar Based Distributed Shared Memory Switch

CXL over Ethernet: A Novel FPGA-based Memory Disaggregation Design in Data Centers

High Throughput Memory Data-Path Design For Multi-Core Architecture

Design and Implementation of a Fibre Channel Target Driver Supporting SCSI

A low‐latency memory‐cube network with dual diagonal mesh topology and bypassed pipelines