Abstract
A robustness enhancement scheme in Serial RapidIO (SRIO) interconnect is proposed to overcome performance degradation caused by noise and Electromagnetic Interference (EMI). The core of this scheme is the adaptive speed transition and mode conversion. Adaptive speed transition improves average throughput and reduces latency in high-noise environments by dynamically adjusting baud rates and feedback intensity based on real-time error rates.
Introduction
Serial RapidIO (SRIO) is a high-performance, packet-switched interconnect widely used for chip-to-chip and board-to-board communication in embedded systems. As data rates reach the Gigabyte-per-second range (1.25Gbps to 3.125Gbps), signals become increasingly sensitive to noise and Electromagnetic Interference (EMI). In environments with high Bit Error Rates (BER), standard SRIO implementations suffer from excessive packet loss and retransmissions, leading to network congestion and performance collapse. This paper introduces the Robust Transport Protocol (ROTP) to handle these lossy behaviors gracefully without sacrificing overall system performance.
Methodology: The Robust Transport Protocol (ROTP)
The ROTP adds a layer of intelligence to the SRIO hierarchy through two primary mechanisms: adaptive speed transitions and mode switching.
Adaptive Speed Transition
Instead of a fixed frequency, ROTP defines three speed states: Low Speed (1.25 Gbps), High Speed (2.5 Gbps), and Full Speed (3.125 Gbps). A state machine manages transitions between these states based on the reception of ACKs and NACKs.
The system "upshifts" to a higher speed after a threshold of consecutive successful ACKs and "downshifts" immediately upon encountering repeated losses (NACKs). This allows the interconnect to maximize throughput when conditions are clear and maintain stability when noise levels rise.
Mode Conversion
ROTP toggles between two operational modes:
- Optimistic Mode (O-mode): Uses sparse feedback to minimize overhead, assuming a stable channel.
- Reliable Mode (R-mode): Uses intensive feedback, acknowledging every packet to ensure data integrity in unstable conditions.
These modes are orthogonal to speed states, allowing for a two-dimensional control space. To implement this within the SRIO standard, the authors extended SRIO control symbols to include status and request functions for speed and mode changes.
Results and Discussion
The scheme was evaluated using Opnet simulations, modeling a communication fabric with varying BER from 0 to $10^{-4}$.
Throughput and Threshold Sensitivity
A critical finding was the sensitivity of the "upshift" threshold. A low threshold makes the system agile but potentially unstable, while a high threshold increases robustness at the cost of slower recovery when conditions improve.
As shown in the simulation results, the adaptive scheme maintains significantly higher average speeds compared to fixed-rate systems in high BER environments. By downshifting to a more stable speed, the system avoids the "retransmission storm" that typically causes massive spikes in latency and throughput degradation.
Conclusion
The proposed ROTP scheme effectively enhances SRIO robustness through dynamic adaptation. Simulation results demonstrate improved throughput and reduced latency in noisy environments. Future work will focus on physical hardware implementation to evaluate power consumption and logic overhead in real-world scenarios.
Summary updated