In industrial automation applications such as cold rolling and hot rolling, extremely stringent requirements govern pressure and speed. Data must be transmitted to the next node control system within a very short timeframe for real-time, precise parameter adjustments. Even minute delays can lead to production errors and significant waste. Currently, the performance of fieldbus and Ethernet technologies cannot fully meet the demands of these highly deterministic and time-sensitive applications. GE's Intelligent Platform's reflective memory technology, designed for demanding simulations, process control, and data acquisition, precisely addresses this deficiency. Its deterministic, low-latency, and high-speed memory sharing features perfectly meet the real-time requirements of such environments.
Low-latency reflective memory is a high-speed network based on a 2.12G baud rate, with a maximum transmission rate of 174Mbyte/s. Shao Jianfeng, an embedded systems application engineer at GE Smart Platform, explains that in a reflective memory fiber optic ring network, when data is written to a local reflective memory backup, high-speed logic synchronously transmits it to the next node in the ring network, and data can be inserted at any node at any time. Each node stores data received from the previous node, unpacks the data packet, checks for errors, writes the new data to its local backup, and then sends it to the next node in the ring network. When the data packet completes its data flow loop and returns to the initial node, it is removed from the network. In this way, each computer always has the latest local backup of the shared memory set, with no software latency and negligible hardware latency. As shown in Figure 1, all computers in the four nodes can receive data written to the reflective memory in just 2.1μs. This low-latency performance plays a crucial role in building real-time systems (such as simulators, PLC controller systems, test benches, or high-availability systems).
As a special type of shared memory system, all CPU accesses to write to this shared memory space are replicated to other nodes on the network, allowing up to 256 computers to be connected. Each subsystem enjoys full and unrestricted access rights.
Besides ring topologies, a star topology for reflected memory data exchange is a developing trend that offers higher synchronization. By using fiber optic hubs, even if a node has been shut down due to an interruption, the hub automatically bypasses any network nodes that have terminated their operations and continues transmission. Each networked computer always has a local, up-to-date backup of the network shared memory set. Backup nodes can seamlessly take over processing work by using the same memory as the failed system, reducing the potentially severe impact of unexpected downtime on productivity, profitability, and performance.
Can it be replaced ? GE reflective memory provides data insertion into a ring network at a rate of 2.12 Gbaud over fiber optic cables, with node distances reaching 10 km (single-mode) / 300 m (multi-mode). Shao Jianfeng said, "Compared to 100 Mbps and Gigabit Ethernet, reflective memory has higher real-time performance, with the longest latency between two adjacent nodes not exceeding 750 nanoseconds. Ethernet and fieldbus currently cannot meet this requirement, and Gigabit Ethernet (including the protocol header) only has a rate of 100 MB/s." (See Figure 2) Due to limitations such as IP protocol costs, addressing, and memory write cycles, achieving the same latency using other network technologies such as Ethernet is also quite difficult.
So, will the development of 10 Gigabit Ethernet in the industrial sector impact reflective memory networks? Shao Jianfeng believes that currently, industrial applications are still mostly limited to 100 Mbps and 1 Gigabit Ethernet, with 10 Gigabit applications remaining at the server level. Large-scale application of 10 Gigabit industrial Ethernet in industrial settings is possible in the long run, but it will still take a considerable amount of time. GE is currently actively developing fiber optic technology with speeds reaching 10 Gigabit to further meet the more demanding real-time requirements of various environments.
Furthermore, as copyrighted by CONTROLENGINEERINGChina (www.cechina.cn), reflective memory does not rely on any network protocol technology, ensuring adequate connectivity without additional load limitations or terminal rules—a significant advantage. "Reflective memory hardware can be used with VME, PCI/PCI-X, PMC, PCIExpress, and various other formats, allowing individual reflective memory networks to connect to different buses. Design and implementation don't need to consider many system compatibility issues, enabling the creation of adaptive systems and facilitating field system setup and expansion," said Shao Jianfeng. "Additionally, protocols mean CPU overhead, and data loss during transmission. Reflective memory networks can transmit raw data with extremely low latency, higher deterministic data transmission, and lower CPU overhead." Reflective memory transparently monitors and replicates data, enabling data sharing without software overhead. Compared to the additional development time, testing, maintenance, documentation, and CPU requirements of traditional communication methods, it is more economical and efficient.
Given its numerous advantages, is reflective memory suitable for all industrial applications? The answer is no. Reflective memory technology is GE's unique real-time networking technology. It can be integrated with other GE embedded platforms to build real-time systems, enabling remote real-time data transmission. It's applicable to all applications that connect computers or programmable logic controllers using Ethernet, Fibre Channel, or other serial networks, such as real-time flight simulators, telecommunications, high-speed process control (steel mills and aluminum plants), and high-speed test and measurement systems. However, it's not suitable for all applications. "Reflective memory is most closely related to systems where real-time interaction is a primary concern. In systems requiring low latency and high communication, although reflective memory boards are more expensive than lower-performance hardware, they offer significant returns through superior functionality and ease of use." Of course, Shao Jianfeng also stated that for some industrial applications with less stringent real-time requirements, and considering production costs, traditional control methods can achieve the desired production goals. Therefore, "reflective memory is particularly suitable for applications requiring high-speed, real-time, and deterministic data transmission. Data ring networks can support up to 256 nodes, fully meeting the application needs of industrial and simulation environments." (PCI5565)
Real-time, precise control via reflective memory has become the optimal solution for many demanding real-time applications in metallurgy, steel, and transportation due to its low latency and deterministic nature. For example, it's used to improve the operational performance of PLC-controlled aluminum rolling mills and control rapid steel rolling processes. On a 3500 ft/min aluminum rolling mill, the response time of a standard PLC control loop allows 2 to 3 feet of aluminum to pass through before actuators can respond, applying and releasing pressure to achieve varying aluminum thicknesses. By utilizing reflective memory, mill-related data is input to the PLC, which rapidly writes the data into the reflective memory, thereby sending the data to a separate VME computer system to transmit complex control algorithms. The system sends output control data back to the PLC via simple reflective memory write commands; the data transmission and calculation speeds are so fast that there is no delay in the PLC's control loop operation. "Delays can prevent the steel from meeting required thicknesses, resulting in wasted costs for certain dimensions. Reflective memory based on the VME advanced control system ensures real-time, precise control, reducing response time to within 4 inches and improving product quality." (PCI-5565)
Shao Jianfeng stated, "Compared to PLC systems, reflective memory offers better real-time performance and faster system response, although it is more expensive. However, the two are not mutually exclusive. Reflective memory complements traditional PLC control, and the choice between them depends on actual production requirements. Furthermore, for system redundancy, reflective memory can restore data from stopped machines to backups within microseconds, a feature also widely used in PLC control." (Reflective Memory Network)
