Abstract: This paper explores and studies the architecture and key technologies of "transparent factory" industrial Ethernet based on Beowulf clusters, providing a feasible and practical solution for the research and construction of "transparent factory" industrial Ethernet, which has guiding significance for related research. Keywords: Beowulf, cluster, transparent factory, industrial Ethernet 1. Introduction Ethernet is a successful information network technology that has been on the market for nearly 30 years and has been widely used in office automation. Ethernet has many advantages such as low cost, stability and reliability, and has become one of the most popular communication networks. The typical communication characteristics of Ethernet are random access, carrier sense, collision detection and contention. However, Ethernet is designed for fields with low real-time requirements such as office automation, and the transmission of messages has the defect of uncertain queuing delay, which cannot guarantee the real-time requirements of data transmission, so it cannot be effectively used in industrial control. With the development of integrated management and control technology, there is an urgent need to extend the Ethernet of the management layer down to the device layer to form a unified transparent data link without data conversion. Therefore, industrial Ethernet technology has become a focus of attention. This paper studies the construction of a "transparent factory" type industrial Ethernet based on the Beowulf cluster, providing a simple and convenient solution for building industrial Ethernet with high performance, low cost, high data processing capabilities, and real-time performance. 2. "Transparent Factory" Type Industrial Ethernet 2.1 "Transparent Factory" and its Characteristics The transparent factory is a creative strategy proposed in the 21st century, integrating technologies that build the internet infrastructure with Schneider Automation's products. Its aim is to bring existing, open, and proven technologies to the market, thereby reducing integration barriers for proprietary systems. The implementation of a transparent factory will significantly reduce the workload of system integration and the time from design and concept to actual product operation. The "transparent factory" is built on new information and communication technologies, including OPC openness, software standards, and Web technologies. A significant feature of the transparent factory is the extension of Web access functionality to the I/O level, enabling remote I/O mediation capabilities, rather than simply accessing I/O information. 2.2 Analysis of "Transparent Factory" Industrial Ethernet "Transparent factory" industrial Ethernet is a complete solution for industrial Ethernet, meaning the enterprise network is entirely composed of Ethernet from top to bottom. It achieves complete "transparency" of the entire factory to the user. Figure 1 compares traditional hybrid industrial Ethernet and "transparent factory" industrial Ethernet. The traditional hybrid industrial Ethernet (Figure 1, left a) is actually a compromise and relatively easy to implement, and is the current industrial network architecture model for many companies. Ethernet is used at the management and workshop levels in the industrial field, connecting upwards to the Internet and downwards to read the status information of various control subsystems in the workshop. At the device level, independent fieldbus technologies such as CAN, Profibus, and ASI are still used. They are connected to Ethernet through dedicated gateways. This solution maximizes the communication capabilities provided by existing Ethernet. Each gateway is an Ethernet node and also the master station of the subsystem below it. The gateway can achieve real-time control of the subsystem and preprocess the information inside the subsystem, sending only important information to the Ethernet in TCP/IP frame format. The advantage of this approach is that it avoids direct Ethernet access to low-level devices such as sensors and actuators, reducing the amount of real-time data transmitted over Ethernet, minimizing the chance of collisions, and mitigating delays in real-time data transmission under harsh industrial conditions to some extent. However, this model is still a transitional structure because the numerous protocols of the fieldbus also affect end-user usage, and the unified and open concept of fieldbus has not been fully realized. Furthermore, in industrial environments, the reliability and data processing capabilities of Ethernet are affected and reduced to some extent, and electromagnetic compatibility needs further improvement. The network architecture model based on the new "transparent factory" industrial Ethernet is shown in Figure 1 (right side b). Its entire network is based on a unified protocol, allowing users to access any device from the high-level management layer to the low-level device layer without needing to know how this access is performed. Ordinary sensors and actuators connect to the Ethernet via industrial Ethernet I/O modules, and the management layer host can communicate with each device via IP access. The advantage of this approach is that it provides a comprehensive solution for field monitoring networks, but many technical challenges still need to be addressed in its implementation. The main modifications to Ethernet need to be made for real-time control, and real-time control algorithms need to be introduced above the network layer to compensate for Ethernet's inherent shortcomings in transmitting deterministic data. Beowulf clustering technology solves the problem of how to achieve high-performance, high-real-time computing systems using ordinary PCs. In recent years, clustering systems have developed rapidly, mainly because the processing power of cluster nodes as workstation systems is becoming increasingly powerful. Faster processors and more efficient multi-CPU machines have entered the market in large numbers, and the introduction of new LAN technologies and protocols has enabled higher bandwidth and lower latency communication between cluster nodes. Clustering systems are also easier to integrate into existing network systems than traditional parallel computers. Clustering development tools are becoming increasingly mature, cheaper, and easier to build. Furthermore, clusters have good scalability, and node performance can be easily improved by increasing memory or improving processor performance. Therefore, using clustering technology to solve the implementation challenges of "transparent factory" industrial Ethernet has become possible. The following discussion explores how to significantly improve the computing processing performance and data transmission capabilities of industrial Ethernet by establishing a Beowulf clustering system, solving key real-time performance issues in industrial Ethernet, and thus realizing the "transparent factory" industrial Ethernet solution. [align=center] (a) Traditional Hybrid Industrial Ethernet (b) "Transparent Factory" Type Industrial Ethernet Figure 1. Comparison of Traditional Hybrid Industrial Ethernet and "Transparent Factory" Type Industrial Ethernet[/align] 3. Beowulf Cluster System A cluster is a loosely coupled multiprocessor system built from a group of independent computer systems. Processes within the system communicate via a network, share memory, and transfer information, thereby achieving distributed parallel computing. A group of inexpensive microcomputers working together can achieve the performance of a supercomputer. Currently, clusters have become a research hotspot for various high-performance computer manufacturers. The definition of Beowulf is still controversial, but most people agree that a low-cost computing cluster built with commercially available microcomputers is called a Beowulf system. The first Beowulf was born in 1994 at NASA's Goddard Space Flight Center. Scientists Sterling and his colleagues connected 16 Intel 486 microcomputers using standard Ethernet to form a computing cluster, achieving a continuous computing power of 70 Mbps. The cluster cost only $40,000 to build, while a supercomputer with the same performance sold on the market at that time cost approximately $1 million. The price advantage of the Beowulf system is unmatched by traditional parallel computers. Besides its powerful computing capabilities and excellent cost-effectiveness, the Beowulf system offers numerous other advantages. First, it relies on mature and readily available computer technologies and equipment, eliminating technical risks. Second, much of the software used to build the Beowulf system can be downloaded freely from the Internet, along with build manuals and help files, reducing costs and simplifying system construction. Third, microcomputers are the most widely used and familiar type of computer to researchers; the Beowulf system, integrated with existing hardware and software, allows users to develop, debug, and run parallel programs in a familiar environment, greatly facilitating user experience. Fourth, the Beowulf system boasts excellent portability and scalability, which are highly beneficial for system construction, maintenance, and resource utilization. 4. Research and Implementation of "Transparent Factory-Type" Industrial Ethernet Based on Beowulf Cluster 4.1 Construction of Beowulf Cluster System PC clusters belong to high-performance cluster computing and parallel computing technologies. The components of a cluster can be adjusted according to the main applications to be run. Therefore, the construction scheme of a cluster system can be adjusted according to specific needs. Beowulf clusters primarily use ordinary, relatively inexpensive computers to build clusters capable of handling heavy computations. Here, we use PC clusters to build the Beowulf cluster system. The Beowulf PC cluster system uses general-purpose PCs as nodes. Each node is tightly integrated from node computers based on Intel and running general-purpose operating systems such as Linux or Windows. It can also have multiple network communication modes or paths; in the industrial Ethernet environment, we use Ethernet communication. [align=center]Figure 2. Schematic diagram of the Beowulf PC cluster system topology[/align] As shown in Figure 2, we use 16 ordinary PCs with PIV-2.66GHz CPUs, 512MB of memory, 80GB/7200RPM hard drives, and dual network cards as nodes to form a cluster, realizing a message-passing-based distributed memory parallel Beowulf PC cluster system. A CISCO2900XL series switch (24 ports/100M) is used, configured with three virtual network segments. One segment is designated as the information receiving segment (LAN1), and the other as the information sending segment (LAN2). LAN1 is only responsible for receiving messages from the node computers and sending the received messages to the various ports of LAN2. LAN2 then sends the messages to the corresponding node computers, allowing each node computer's two network cards to perform message sending and receiving respectively, improving message transmission speed. The third network segment of the switch is used to connect the system to the local area network, thereby enabling functions such as remote login services. To achieve faster operating speeds, unnecessary services in the operating systems of each node can be removed to improve application execution efficiency. Furthermore, NFS communication and data communication can be bound to different network interface cards (NICs) to increase data transmission speed. It's even possible to install three NICs on each node, one bound to NFS and the other two to data communication, achieving high-speed communication. Using the same techniques, a Beowulf system with 48 Athlon 1G nodes can achieve a peak computing speed of 38 Gflops/sec. 4.2 Software Environment Setup The software typically running on a Beowulf cluster includes the Linux operating system, Parallel Virtual Machines (PVMs), and Message Passing Interfaces (MPIs). The message passing model is widely used in cluster environments. While the message passing model provides flexibility for programmers, it also places the burden of complex information exchange and coordination between parallel execution parts on the programmer, increasing their workload—a disadvantage of the message passing programming model. Currently, popular message passing software includes Parallel Virtual Machines (PVMs) and Message Passing Interfaces (MPIs). MPI is an explicit message-passing model where tasks communicate with each other by sending messages. Its biggest advantage is high performance; its point-to-point communication function model and operable data types are richer than PVM, and its group communication function library is also larger, but it is less flexible than PVM. Both MPI and PVM provide a set of functions, each with its own specialization. They can run on all parallel platforms, including PVP, SMP, MPP (massively parallel processor), workstations, and PC cluster systems, and have been implemented on the Windows platform, providing bindings for C, Fortran, and Java languages. Beowulf cluster systems typically use UNIX, Linux, or Windows NT/2000/XP operating systems. These all have strong network support and reliability. Linux, as a well-known open-source operating system, can be downloaded for free on the Internet and is used by most Beowulf systems. In the actual project implementation, Red Hat Linux 9.0 was used. After installing Linux, network configuration was performed. TCP/IP network protocol was installed on all node microcomputers, and all node microcomputers were set to the same workgroup, with different network names defined for differentiation. Next, install the MPI software MPICH 1.2.5.2 on the node microcomputers as the supporting environment for the cluster, and set the hard disk partition or folder storing the parallel program to be shared. 4.3 Implementation of Beowulf Cluster System in "Transparent Factory" Industrial Ethernet [align=center] Figure 3, Topology Diagram of "Transparent Factory" Industrial Ethernet Based on Beowulf Cluster[/align] The architecture of "Transparent Factory" Industrial Ethernet based on Beowulf cluster is shown in Figure 3. In the new industrial Ethernet, the Beowulf cluster system is connected to the central switch like a general-purpose computer. The entire network is divided into four layers: computer management layer (computer center), management layer (management layer), workshop layer (industrial PCs on the production line), and equipment layer. Each layer is connected through a high-speed central switch. In addition, remote management personnel can access the network via the Internet to control the entire "transparent factory". Of course, to realize a truly "transparent factory" industrial Ethernet, it is also necessary to solve the security problems of the CSMA/CD eavesdropping mechanism, data conflict problems, and other related issues, which will not be discussed here due to space limitations. 5. Performance Testing and Result Analysis Industrial Ethernet testing typically includes the following important indicators: (1) Maximum response time. For each node's message, under normal network operation, the maximum delay time experienced from sending to receiving and generating a response. (2) Packet loss rate. Under normal operation, the number of packets lost within a specified time. (3) Transmission medium. The test results of ordinary Ethernet transmission media in industrial fields are not satisfactory. Whether the Ethernet transmission medium can guarantee stable and reliable data transmission is an important selection criterion. (4) Transmission rate. Under normal network operation, the number of bits that can be transmitted per unit time. The standard test conditions for dynamic traffic filtering industrial Ethernet are: frame length 512 bytes, test time 600s, and frame interval 9.6μs. The test results are as follows: traffic: 300MB/s, maximum response time: 187.3ns; number of lost frames: 11. As can be seen from the test results, in industrial applications where real-time requirements are not extremely high, this solution has been completed and can meet the requirements. The above solution achieved nearly 80% cost savings in the implementation of industrial control computer management and control in solar cell production lines, while also achieving excellent results. 6. Conclusion and Outlook This paper discusses how to construct a "transparent factory-type" industrial Ethernet based on a Beowulf cluster. This system adopts a distributed, homogeneous, peer-to-peer cluster structure composed of microcomputers and high-speed Ethernet networks. The Beowulf cluster system, consisting of 16 microcomputers, serves as the computing and data processing center within the Ethernet network, achieving the specifications of a high-performance supercomputer. Ethernet, due to its simple structure and ease of construction, can achieve high LAN data transmission performance at a low cost, thus complementing the Beowulf cluster system and jointly solving the real-time data transmission problem in industrial Ethernet. References: 1. Korch M, Rauber T. Evaluation of Task Pools for the Implementation of Parallel Irregular Algorithms[C]. Proceedings of the International Conference on Parallel Processing Workshops, 2002. 2. Li Guiming, Yu Guoyang, Luo Jiarong. Implementation of Beowulf Cluster Based on Linux[J]. Computer Engineering. 2003: 29(II). 3. Wang Tao, Li Xiaoming. SMP Cluster: How to Develop Two-Level Parallelism[J]. Computer Engineering and Science. 2002, 24(4). 4. Li Chong, Luo Jiarong, Wang Huazhong. Design and Research of PC Cluster System Based on BEOWULF and Parallel Programming[J]. Microcomputer Information. 2005, 21(8-3): 64-67