Finding a slow plug-in in ALTRON with the Performance and Internal Info tabs

Frames are dropping, the actual rate is below the requested number and your processing graph has ten or twelve nodes. Which one is to blame? To find a slow plug-in in ALTRON, you don't need to guess or remove nodes one by one. Two small tabs in the diagnostics panel put the answer on screen.

The processing graph in ALTRON is a chain of nodes. Each node is a plug-in that takes data from the previous node and passes it to the next. If one of these links is slow, everything behind it waits.

Step 1: overall CPU usage

Below the list of detected cameras you see three indicators: Physical RAM, Virtual RAM and CPU. Each shows total system usage and the program's own usage separately. If CPU is near 100 percent, dropped frames are no surprise. Now you need to see who is using the CPU.

Step 2: the Performance tab

This tab is a simple table. The Cpu column shows each node's share of the CPU. The node's name appears in the Item column. The table refreshes every second. There is also a Ui row, which shows the share of the user interface itself. If several cameras are open at once, each chain's rows appear under a separate heading labeled Chain with the chain number.

The method we use when troubleshooting:

1. Start acquisition with exactly the settings where the problem appeared.
2. Open the Performance tab and wait a few seconds for the numbers to settle.
3. Find the row whose Cpu value is clearly larger than the rest.
4. Make one change in that node and look at the table again. Only one change at a time.

Step 3: Internal Info to see where data is dropped

The problem isn't always the CPU. Sometimes a node with normal CPU usage leaves data behind. That is what the Internal Info tab is for:

• Count Rate: It shows how many packets pass through each input and output port of each node every second.
• Main / Extra Data Rate: Main shows the main data rate in megabytes per second. Extra is the rate of the additional bytes that travel with the data.

The rule is simple: if a node's output rate suddenly drops below its input rate, data is being dropped right there. Raw data saving nodes also report their disk write rate in this tab. And if you see no image at all, first check whether the Count Rate of the camera's output port is zero.

You've found the slow node. Now what?

• Make its settings lighter. For example, in the color correction matrix plug-in, make sure Use Lookup Table is on. That is the default, and turning it off uses more CPU. Another example is saving demosaiced raw data: the higher you set Compression, the smaller the file, but the heavier the processing load.
• Put an ImagingInterval in Max rate mode in front of it. This node caps the data throughput and discards any data that arrives above the cap. In the ready-made graphs you will see it named DREG. Its usual place is right before the heavy node or the display.
• Check Read Buffer Size. The larger the buffer, the less often the USB driver is called and the lower the CPU load. If someone has lowered it, set it back to a larger value.
• Remove unneeded nodes. Every node whose result you don't need takes a share of the CPU.

A mistake we see often

ImagingInterval in Max rate mode discards extra data. If you put it at the start of the chain, before the saving branch, the recorded file will have that same low rate. When you need every single frame, put this node only in the display branch or in front of a heavy node that doesn't need every frame.

One small tip from experience: read the Cpu value in steady state, not in the first few seconds after Start. At that moment everything is still starting up and the numbers are misleading.

For a more general explanation of software profiling, see Wikipedia (English).

If you can't interpret the Performance table for your graph, send us a screenshot of it through contact form.