As industrial and other safety-critical systems consolidate functions onto shared computing platforms, communication between those functions becomes a design concern in its own right. A real-time virtual switch can provide a controlled path for messages between software components, while helping preserve timing behavior and separation between data flows. Combining software-defined networking (SDN) with Time-Sensitive Networking (TSN) offers one approach to making that switching both configurable and time-aware.
The central challenge is not simply moving packets. Integrated systems may host functions with different criticality levels, and communication must avoid unintended interference while meeting timing requirements. A proposed virtual switch architecture addresses this challenge using IEEE 802.1Qbv and IEEE 802.1Qci mechanisms, with a proof-of-concept implementation on a real-time Linux platform.
Why integrated systems need predictable communication
In a federated architecture, functions are assigned to separate computing nodes. Integration brings multiple functions onto a shared platform, potentially reducing duplicated hardware, but it also makes isolation essential. A fault or resource conflict affecting one function should not propagate to another, particularly when the applications have different safety or criticality requirements.
Standards and platforms such as ARINC 653 and AUTOSAR provide ways to structure applications and their communication. However, communication mechanisms in these environments commonly rely on static, offline configurations. That can make changes during operation difficult to accommodate. Meanwhile, general-purpose software switches such as Open vSwitch support programmable forwarding, but their conventional design does not by itself address the bounded latency and low jitter required by hard real-time communication.
For a dependable communication service, a switch must therefore address more than routing. It needs timing control at both ingress and egress, spatial separation between flows, and a way to update configuration without abandoning deterministic behavior.
How a real-time virtual switch uses TSN
The proposed design draws on SDN’s separation of the control plane from the data-forwarding plane. Forwarding behavior can be changed by updating configuration, while the switch handles message processing according to that configuration. TSN mechanisms provide timing controls for selected traffic.
Ingress checks with IEEE 802.1Qci
For safety-critical, time-triggered traffic, each message is expected to arrive within an assigned time slot. The virtual switch checks the message’s arrival time against a configured, time-based access control list (ACL). Messages that arrive at the expected time can proceed to the queues for their destination egress ports. Messages that arrive outside their permitted window are treated as untimely and discarded.
This ingress check helps prevent out-of-window traffic from entering the scheduled forwarding path. The switch also uses routing information to identify the appropriate egress port or ports for a message.
Egress scheduling with IEEE 802.1Qbv
At egress, a Gate Control List (GCL) determines when queues may transmit. The schedule divides a repeating cycle into intervals, each associated with a queue mask that specifies which queues are enabled. For transmission to proceed, the relevant queue must be open and enough time must remain for the message to be sent; otherwise, the message stays buffered for a later opportunity.
The ingress ACL and egress GCL must be coordinated, including the switch’s relay overhead. Queues also need to be arranged so messages from different flows do not contend in a way that undermines the intended timing. The design assigns dedicated queues to critical data flows, while egress access is controlled so that messages use a port exclusively at a given transmission time.
Isolation and configuration on a shared platform
The architecture models virtual end systems communicating either through one virtual switch or through multiple virtual switches on the same platform. In both cases, switching is incorporated into a time-triggered schedule alongside the application tasks that produce and consume messages. This makes the communication schedule part of the wider system design rather than an independent networking configuration.
In the proof of concept, each virtual switch was assigned a dedicated physical processor core. The implementation used a six-core PC with a 3.2 GHz processor and 16 GB of RAM, running Linux LXRT/RTAI. Real-time FIFOs provided communication channels between virtual end systems and switches, and between switches. Shared memory held configuration such as routing information and GCLs, allowing a manager process to provide current settings to the virtual switches.
The implementation used static data structures for GCLs and per-port queues. This was intended to support predictable processing behavior. Separate ingress processes were scheduled in a 10 ms period, each with a 5 ms runtime, to provide temporal separation within the virtual switch. These choices describe the specific prototype, not universal requirements for every implementation.
What the experimental results show
Tests measured the time consumed by switching between virtual end systems. For local communication through one virtual switch, most reported switching overhead fell between 0.5 and 2 microseconds, with some observed exceptions reaching 5 microseconds. The authors associated these variations with factors including timer interrupts handled on other cores and jitter from the RTAI scheduler.
For a message passing through two virtual switches, most of the measured overhead was around 5,000 to 5,005 microseconds. The approximately 5 ms component reflected the configured process schedule: the second switch received messages after the first switch’s dispatching process. The remaining variation was attributed to implementation effects, including inter-core communication and scheduler jitter. These measurements demonstrate behavior in the stated setup; they should not be read as a general latency guarantee for other hardware, schedules, or workloads.
Practical implications for system designers
A TSN-enabled virtual switch offers a framework for combining configurable forwarding with scheduled, isolated traffic handling inside an integrated computing platform. For engineers and system integrators, the main design considerations include:
- Align schedules end to end: Coordinate task timing, ingress windows, relay time, and egress gates.
- Plan queue and flow assignments: Ensure critical flows have the intended separation at shared egress ports.
- Account for platform effects: Core allocation, timer handling, and real-time scheduling can influence observed jitter.
- Manage configuration carefully: Dynamic updates must remain consistent across the switches and the wider system.
The demonstrated system ran on a single computing node. Extending the approach to environments that combine virtual and physical switches remains an additional challenge, particularly when coordinating configuration across the complete network.
Conclusion
Integrating applications onto shared hardware makes dependable internal communication as important as application partitioning. A real-time virtual switch using IEEE 802.1Qci ingress checks and IEEE 802.1Qbv egress scheduling provides a way to control when critical messages enter and leave the switching path, while SDN principles support configurable forwarding. The prototype results show that deterministic switching can be demonstrated in a dedicated real-time execution environment, while also highlighting the need to account for scheduling delays, jitter, and system-wide configuration when applying the approach.