Making 10GigE machine vision reliable and affordable - including multi-camera setups

Making 10GigE machine vision reliable and affordable - including multi-camera setups

Overview

A variety of vision system application scenarios such as fast inspection lines, semiconductor factories, intelligent transportation systems, motion analysis and volumetric capture require high resolution, high FPS and high data transfer rate to achieve better results. For vision system engineers who want to improve the output by taking advantage of machine vision cameras with faster frame rates and higher resolution, upgrading from 1GigE to 10GigE is an obvious choice. However, according to research by the AIA (Automated Imaging Association), its adoption has been quite slow. This is understandable considering the three technical challenges that come with this upgrade: reliability (packet loss), high CPU utilization and high latency. This article describes how the Teledyne FLIR Oryx + Myricom bundled solution addresses these challenges.

Update 1: Perfect performance

While 10GigE Vision offers 10 times higher bandwidth than the GigE Vision protocol, 10GigE host adapter performance has not improved accordingly. Data transfer from the camera to the host often results in CPU overload, application buffer overflows, and unacceptable levels of packet loss for demanding applications.

By leveraging the host adapter to handle packet reception and image reconstruction directly on the card, the CPU no longer needs to manage these tasks. The Teledyne FLIR Oryx + Myricom bundled solution is designed to handle this scenario. As shown in our test results below, system reliability can be greatly improved, resulting in significantly less packet loss and, in turn, less frame loss.

This bundled solution works seamlessly with our new custom SDK driver specifically designed to process the data provided by the Myricom card. This combination allows for a flawless and reliable transfer of image data from the camera to the host PC. The results are shown in the Appendix below: Reliability and CPU Utilization Tests.

The price/performance ratio of the Teledyne FLIR Oryx + Myricom bundled solution made it an obvious choice; compared to purchasing the hardware separately and integrating it, it is an affordable and highly reliable setup.

Update 2: CPU usage is manageable

Theoretically, the CPU can use up to 100% of one core to process the incoming data from the 10GigE connection, and multiple cores can be used when running multiple applications/cameras. By using the Myricom card to manage packet reception and image reconstruction, the CPU utilization of each application can be as low as 1%, allowing more CPU cycles to be used for image processing. The test results are shown in the appendix below: Reliability and CPU utilization test

Update 3: Delay reduced

10GigE Vision frame latency is not deterministic; this means that frames can arrive with significant timing jitter. In some cases, especially with switches, not only are packets lost, but frames are sometimes received out of order. The Teledyne FLIR Oryx + Myricom bundled solution solves this problem by providing timely notification of frame completion to reduce latency and reduce timing jitter.

Appendix: Reliability and CPU usage testing

Test 1: 7-day streaming at high bandwidth

Using a custom console application created with the Teledyne FLIR Spinnaker API, an 8.9-megapixel Teledyne FLIR Oryx camera was set up to continuously capture images and keep track of any incomplete images, with no additional processing or third-party resource-intensive programs running simultaneously.

Test results: About 40 million frames of images were collected; 0 incomplete/lost frames were detected.

Note: CPU usage was checked throughout the 7-day test period and was consistently at 1%. With the new Myricom driver disabled and relying solely on the FLIR standard filter driver, CPU usage remained at approximately 100% on the CPU cores dedicated to the application.

Test 2: Dual camera streaming

The test involved two Oryx cameras (ORX-10G-123S6M and ORX-10G-89S6C) running in the same custom console application, with a bandwidth of 6.7 Gb/s per camera, for 24 hours.

Test results: Each camera collected about 6 million frames of images; 0 incomplete/lost frames were detected

Test 3: 24-hour CPU stress test

This test included an Oryx camera (ORX-10G-123S6M) with the same setup as Test 1.

The same console application as in Test 1 was used, but this time a different application was used simultaneously; this custom application was designed to simulate a heavy workload with total CPU utilization of approximately 90% (all eight cores).

Test results: about 6 million frames of images were collected; 0 incomplete/lost frames were detected