Improving Multi Camera Use for GigE Vision systems
3 Aug 2022 | Reading time: 3 min
When it comes to interfaces in the machine vision industry, users face a variety of options to choose from to fit their application best. Thereby, every interface has unique pros and cons.
The Ethernet based GigE Vision standard is still considered the most frequently used interface in machine vision. It offers a large variety of compatible standard network equipment and enables scaling the performance of the whole system over a wide range. This makes GigE Vision especially suitable for multi-camera systems.
However, designing reliable high-performance machine vision systems using multiple cameras usually requires considerable experience and effort. The proper selection of system components and design decisions can be a time-consuming task. Balancing cost, reliability and performance may require multiple design cycles with application specific tests.
GigE Vision and its limitations
GigE Vision sits on top of very common network technology: its core protocols GigE Vision Control Protocol (GVCP) and GigE Vision Streaming Protocol (GVSP) are mainly based on the User Datagram Protocol (UDP).
It allows efficient and low latency transmission of data. However, UDP has some drawbacks that need to be considered:
Inability to detect and correct data which got lost on the way
No guarantee of the right order of datagrams
Susceptible to network congestion or system overload
The order and comprehensiveness of data is ensured through the GigE Vision protocol. The device sends additional information in each stream packet header (e.g., frame number). This allows the host to check, whether packets or complete frames are missing. If the host determines that packets have been missed, it may send a request to the camera to repeat the transmission of the missing packets. However, this beneficial function can temporarily increase the system load, thereby harming the performance of the overall system.
System overload in GigE Vision Multi-Camera Setup
In GigE Vision systems, overload is typically characterised by the loss of packets. Unfortunately, the packed-resend mechanism may cause a self-amplifying effect: lost packets create resend requests, which increase the system load, and increases the probability of packet loss. The result is system hiccups with many dropped frames.
This issue can be avoided by reserving enough throughput margin during the system design. But how much margin is required? This depends on the requirements, system performance, throughput, and network topology.
Usually, considerable testing efforts and experience are required to get a well-balanced and cost-effective system. Multi-camera setups combined with high throughput requirements need to be developed meticulously to avoid bottlenecks and reliability issues.
Ethernet Flow Control – Relieving the host system
A typical cause for packet loss in overloaded systems is the overflow of a packet buffer in a network device. These hardware buffers are found in network switches and network interface cards (NICs). They are usually able to buffer only milliseconds of full throughput, which means that a small, temporary congestion event may lead to packet loss.
The Ethernet flow control feature, implemented in Allied Vision’s Alvium G1 & Alvium G5 camera series, can avoid the overflow of these buffers. The receiver can send a pause request to the transmitter if it detects a nearly full buffer. Then, the transmitter stops transmission with the next Ethernet frame for the requested amount of time or until it receives a request from the receiver to continue.
Meanwhile, the camera's internal memory, which usually is significantly larger than network device buffers, accumulates the data which arrives from the image sensor and holds it back.
As long as the mean throughput allows enough room to catch up the temporary backlog from the camera's buffer, Alvium GigE cameras ensure a reliable data transmission and avoid packet loss in the network.
Easy to use
Flow Control will be enabled by default in all Alvium based GigE Vision cameras and almost every Network Switch and Network Interface Card supports this feature. Users only must make sure the feature is also enabled on the network device side.
Packet loss due to congestion is a typical issue in high throughput GigE Vision based machine vision systems. The Ethernet flow control mechanism helps to improve the utilisation of buffers throughout the whole system so that it can absorb temporary congestion events much better.
Consequently, it is much easier to design a reliable high throughput system within the limitations of technology, budget and time.
Author: Maximilian Poggensee
Product Manager Allied Vision Technologies GmbH