Enabling Ultra-High-Speed and High-Resolution Machine Vision Applications
Myricom’s MVA™ software, running on Myri-10G extreme performance 10-Gigabit Ethernet adapters, provides a high performance platform for GigE Vision®-enabled devices for machine vision applications like mva cameras for video surveillance, high-speed industrial imaging and quality control inspection within medical, semiconductor fabrication, food & beverage, and automotive industries.
Myricom MVA™ enables the lowest host-CPU network processing overhead and latency available for Ethernet-based machine vision applications. MVA enables applications to receive images from high-speed devices such as cameras, while maximizing the CPU cycles available for image analysis. MVA typically consumes less than 5% of one CPU core to receive 2MB frames at 10GbE line rate. A single MVA-equipped server can support 8 or more 10GigE Vision devices, or up to 64 GigE Vision devices. It aggregates input from more high-performance devices into a single server than was previously possible.
MVA accelerates GigE Vision 1.2 compliant stacks. The GigE Vision protocol has become one of the dominant standards in machine vision by leveraging the reliability, long cable length, flexible network topology, performance and low cost of existing Ethernet infrastructures. Despite these advantages, latency and host-CPU overhead have limited system performance, particularly compared against other protocols such as FireWire and Camera Link.
Now, MVA enables low-latency and low-CPU overhead GigE Vision machine vision applications. MVA-enabled Myri-10G adapters offload GigE Vision Stream Protocol (GVSP) which allows the transfer of image data directly to the application’s frame buffers without additional copies or host CPU involvement. This bypass eliminates intermediate copying of packets that comprises the bulk of the CPU overhead in packet processing.
Further reduction of CPU overhead and latency are achieved with Myricom MVA’s intelligent interrupt moderation where a single interrupt is generated per frame. Network adapters that are not GVSP-aware typically generate many interrupts per frame to reduce latency, but do so at the cost of higher CPU usage.
Key Features and Benefits
- Low CPU overhead: Dramatic reduction in CPU utilization means more CPU cycles for image processing.
- Lower latency: Timely notification of frame completions means lower latency and less jitter.
- Fewer dropped frames: Lower CPU overhead means higher throughput and more responsive applications, leading to fewer dropped frames due to application buffer overflows.
- Interoperability with Ethernet driver: Multiple GigE Vision applications can receive images from multiple devices while the standard Ethernet driver is still available for non-GigE Vision traffic.
- CPU load balancing: De-multiplexing of GVSP at the Myri-10G adapter allows each imaging device to be steered to the appropriate buffer for each application.