Network performance expectations

One of the comments that tend to pop up about ManT1S is: "Why resurrect 10 Mbit/s networking, when we now have multi-gigabit alternatives?" Some people like to oversimplify complex engineering decisions to "just go with the highest number" or "I'll settle for nothing but the best", where "the best" invariably is defined as some high specification number in one single dimension.

While that's the kind of story marketing departments may like to push, engineering requires a different approach. The fact is that engineering is always about making decisions in a multidimensional parameter space, about making the right trade-offs for the application, and about choosing the right tool for the job.

Engineering Questions

With this engineering mindset, bandwidth is only a single performance metric among many others. Some examples of questions to determine whether T1S and the ManT1S might be a good fit for an application are:

• What bandwidth do I really need?
• Am I using T1S to replace a communication channel that was previously served with a UART connection or CAN bus? Then 10 Mbit/s bandwidth is not going to be an issue.
• If I was using a higher speed network, did I measure how much bandwidth I actually need? People often overestimate how much bandwidth their application really uses. Even video can work fine with 10 Mbit/s bandwidth.
• Is bandwidth actually important, or is low latency what I really need?
• On the ManT1S, you may get higher bandwidth with Wi-Fi, but you will get significantly better and more consistent latency with T1S (see latency numbers in this guide). This is often much more important for control tasks.
• This also goes into the question of whether you really need a Linux based SBC or if an embedded controller like ESP32 is sufficient or even better for the application because of more consistent latency and real-time behavior.
• How much does my application get simplified by using a multi-drop bus like T1S vs needing point-to-point connections?
• Not needing a switch can be a significant simplification versus possibly faster point-to-point networking, which requires a networking switch.
• The multi-drop bus topology preserves easy expandability if you may need to add nodes in the future versus something like point-to-point communication channels between individual boards.
• How am I going to deal with power?
• Will each node need its own power source or could I take advantage of power over data line to distribute power? Not having to worry about power can be a significant simplification.
• Will I get in trouble with ground loops if I don't distribute power from a single point but create loops of networking and power wiring that take different paths?
• Can I significantly reduce my wiring harnesses by using networked nodes instead? Copper is expensive nowadays, it could simplify the system and lower cost to distribuite the system between several smart nodes.

As you can see, there are many things to consider. ManT1S may or may not be the right fit for your application, but don't fall into the trap of dismissing it just because the bandwidth is 10 Mbit/s. Use cases in embedded systems are plentiful and often don't require high bandwidth. In many cases, CPU speed or memory will limit throughput before networking bandwidth becomes a limitation.

What bandwidth can I actually expect in practice?

10BASE-T1S specified bandwidth is 10 Mbit/s but considering other limitations and overhead, what bandwidth can actually be expected in real use?

I ran Espressif's iperf example code on two ManT1S'es and got this result:

I also tried to run Espressif's iperf on one ManT1S connected to a wESP32 through a ManT1S-Bridge to check a heterogeneous system with T1S and regular Ethernet. I got this result:

In both cases, 9+ Mbit/s shows that the overhead is limited and throughput is close to the theoretical maximum.