In real-time embedded systems like VxWorks, reliable and predictable Ethernet communication is critical. To improve both reliability and real-time performance, we can model the network behavior mathematically and apply queueing theory to determine the optimal buffer size that minimizes both packet loss and latency.
This blog explores the modeling approach, optimization process, and engineering recommendations for VxWorks network performance tuning.
Why Network Modeling Matters in VxWorks #
When Ethernet frames are sent in a VxWorks system, delays can arise from:
- Transmission delay – due to the physical medium.
- Queueing delay – when frames wait in the send buffer before transmission.
If a frame is not transmitted within the current cycle, it must wait until the next one, leading to significant delays. On the other hand, too short a queue can underutilize the link. The goal: find the sweet spot in buffer size.
Building the Ethernet Queue Model #
The frame sending process in VxWorks can be modeled with Poisson arrivals (independent frame arrivals, no limit on count). Let:
- λ – average frame arrival rate (frames/time unit)
- μ – average transmission rate (frames/time unit, μ > λ)
- ρ = λ / μ – traffic intensity
- L – buffer length
- L₀ – optimal buffer length
If L is too large → frames pile up, some get dropped due to delay.
If L is too small → link goes idle, wasting capacity.
The Loss-Cost Optimization Approach #
We assign loss costs:
- c₁ – cost of dropping a frame when
L > L₀ - c₂ – cost of idle capacity when
L < L₀(typically c₁ ≥ c₂)
By deriving steady-state probabilities from the birth-death process and geometric distribution, we can calculate:
-
N_d – average dropped frames when busy
-
N_p – average idle frames when underutilized
-
F(L) – total communication loss cost:
F(L) = c₁ * N_d + c₂ * N_p
The optimal queue length L₀ is the integer that minimizes F(L).
Visualization: Optimal Queue Length #
Below is a conceptual plot showing the loss cost curve versus queue length, with the optimal point L₀ marked.
Visualization: Cost Components #
To better understand the trade-offs, the following diagram separates the dropped-frame cost, idle-capacity cost, and the total loss cost. The intersection of these trade-offs determines the optimal queue length.
Key Engineering Recommendations #
From simulation and analysis, we found that tuning the buffer size significantly reduces queueing delays and improves performance. Beyond the math, here are practical tips for VxWorks network design:
- Design the right topology – For critical nodes, use mesh structures with redundancy.
- Use separate send and receive buffers – Enable zero communication dead time on the application layer.
- Increase resources to avoid congestion – Wider bandwidth, multiple routers, backup links.
- Tune timeout and retransmission – Set timeout equal to one full round trip (send, process, and ACK).
- Separate data and control channels – Use two sockets: one for data, one for monitoring/ack.
- Fast reconnection handling – On link failure, quickly re-establish sockets to resume normal operation.
Conclusion #
By modeling Ethernet queueing behavior in VxWorks and finding the optimal buffer size, we can reduce packet loss, improve utilization, and enhance real-time performance. Combined with thoughtful topology, timeout tuning, and redundancy, these strategies can significantly boost VxWorks network reliability.
This approach has strong engineering applicability for embedded systems requiring deterministic network performance.
References
- Wang G., Yue S., Li Y., et al. Networked Command and Control System Software Architecture Research, Modern Defense Technology, 2013(2).
- Qiu A., Zhang T., Gu Y. Real-Time Ethernet for Spacecraft Applications, Journal of Space Science, 2015(3).
- Miao X. Architecture of 11 Industrial Real-Time Ethernet Standards, Instrumentation Standardization & Measurement, 2009(3).
- Lu C. Queueing Theory, Beijing University of Posts and Telecommunications Press, 2000.