In modern distributed architectures, performance issues rarely stem from a simple lack of bandwidth. Even when network utilization registers at only 20% to 30%, systems can still suffer from dropped packets, unpredictable latency spikes, and silent communication failures.
As illustrated in Figure 1, data flows across complex topologies linking multiple hosts through switches, intermediate routers, and even across the internet cloud – and choke points can hide at every tier. Switch buffers overflow during sub-millisecond microbursts. Faulty transceivers quietly discard corrupted frames. Routers silently swallow oversized packets when firewalls block fragmentation alerts. At the host operating system level, overloaded processor cores drop incoming packets before user applications ever see them. To untangle these issues, both Perftest and Perftest Lite are built to communicate across these different network segments, testing seamlessly across local switches, traversed routers, and wide-area internet connections to identify data flow problems.
Addressing these challenges requires moving away from guesswork and adopting concrete data profiling. Perftest and Perftest Lite provide the dedicated benchmarking capabilities required to stress-test nodes, characterize network architectures, and resolve issues before finalizing your network hardware and architecture.
The Core Capabilities of Perftest
Perftest deploys an end-to-end benchmarking setup featuring a publisher and a subscriber communicating across switches, routers, or wide-area internet connections.
To establish a baseline independent of Data Distribution Service (DDS®), Perftest can transmit raw UDP datagrams configured to reflect expected operational message sizes. This reveals what the physical network infrastructure is natively capable of handling by directly measuring maximum raw data throughput and baseline message latency.
After establishing this baseline, Perftest can characterize the system using the DDS protocol as implemented by RTI Connext. By passing identical message payloads over DDS, the tool records maximum achievable data rates, latency, and the critical metric of latency under load. It also allows you to test different DDS Quality of Service (QoS) configurations to best meet latency and throughput requirements under expected network load.
Expanding to Embedded Systems with Perftest Lite
Standard benchmarking suites often assume high-performance hardware running full operating systems like Linux or Windows. However, real-world edge systems – such as avionics, automotive controllers, and medical devices – frequently run on resource-constrained platforms using Real-Time Operating Systems (RTOS).
Perftest Lite was written in a way to simplify porting to resource-constrained platforms, thus making it easier to compare their performance. Perftest Lite can be integrated with lightweight Connext Micro libraries to bring rigorous stress-testing to embedded environments. Crucially, it provides complete end-to-end benchmarking directly between resource-constrained embedded targets running Perftest Lite on both ends, while also offering the full flexibility to pair with standard Perftest for hybrid, cross-platform evaluations between an RTOS-based node and a full enterprise Linux host (as an example).
Embedded teams can capture maximum data throughput and topic latency directly on low-footprint hardware, providing predictability for mission-critical operations. The resulting diagnostic data allows engineers to fine-tune RTOS memory pools, optimize switch configurations, or calibrate QoS parameters to match embedded operational constraints before going into production.
Try it Out – Characterize Your System !
Access to Perftest or Perftest Lite (depending on your project) is available through source code and can be found in:
Perftest: https://github.com/rticommunity/rtiperftest
Perftest Lite: https://github.com/rticommunity/rtiperftest-lite
If you’d like assistance in accessing, building or using Perftest, feel free to reach out to to send an email to the RTI FAE team, we are here to help.
About the author: