Title

Marya D. Pearson

Office of Science, Science Undergraduate Laboratory Internship Program

StanfordUniversity

Stanford Linear AcceleratorCenter

Menlo Park, California

August 15, 2008

Prepared in partial fulfillment of the requirement of the Office of Science, Department of Energy’s Science Undergraduate Laboratory Internship under the direction of Ernest L. Williams Jr. in the Controls division at StanfordLinearAcceleratorCenter.

Participant:______

Signature

Research Advisor:______

Signature

ABSTRACT

Marya Pearson (Norfolk State University, Norfolk, VA 23504) Ernest Williams Jr. (Stanford Linear AcceleratorCenter, Menlo Park, CA94025).

INTRODUCTION

Feedback is a concept of communication, and in this aspect it is communication between many diagnostic devicesthat require feedback to maintain control of theLinac Coherent Light Source (LCLS) beam line. High energy electrons travel the two-mile linear accelerator (linac) for physics experiments. For quality experimental values, the electron beam must maintain its position at the origin of the transverse plain. In order to allow the beam to consistently travel in the correct position, a network is needed to facilitate communication between the various control and actuator devices.

The network implemented in the linac must satisfy some constraintsoutlined by LCLS physicists in the Physics Requirement Document (PRD). From this document, the fast feedback network should aim to function at a rate of 120 Hz, and providean architecture that is scaleable, flexible, deterministic and reliable. Some network connections exist between the beam position monitor (BPM), a private BPM switch, and designated input/output control (IOC) computers. It is the connection from the IOC, a master CPU, and the actuators that is the focus of this project. In short, the IOC receives BPM data in ≤ 1ms, and then passes the information to a master CPU which creates algorithms with the data in Matlab to produce instructions for the actuators. The actuators, either magnets or klystrons, receive this data in ≤ 1ms and respond to the instructions in 6ms. Thereby, in approximately 8ms, the entire feedback loop is complete. According to the 8.3 ms time budget, this time estimate satisfies the outlined constraint, however, it hardly provides a cushion. By choosing an appropriate network technology and architecture, the 1ms between IOC, CPU, and actuator can be reduced and deliver the fast feedback for LCLS.

METHODS

While there are many network technologies to choose from, the most suitable options for this project are reflective memory, point-to-point Ethernet, and multicast. Of the many technologies, these three feature similar and unique characteristics. Cost, latency, speed, error, and data transfer, along with the mentioned physics requirements, are the decisive factors for comparing these technologies. Reflective memory (RM) is a memory bus technology in which data is replicated and stored in local memory until it reaches its destination [ ]. For example, if the IOC, CPU, and actuator are connected using RM, the data from the IOC would be copied and stored at each node. Though this function can be done speedily, the disadvantage is that every node connected with RM will receive the same data whether it needs the data or not; in this case the actuators would receive unnecessary BPM data. Another consideration for RM is the system topology. Reflective memory can be configured in either a star topology or a ring topology. The two differ in the hardware required for a connection. A star topology is connected using RM cabling with transmit, receive ends, a RM hub that allows up to 256 nodes to connect, and a VMIPMC card that houses the actual RM technology (Refer to Appendix). Then, ring topology simply connects each device directly to the next using RM cabling. For example, a ring configuration between the IOC, CPU, and actuator would attach the IOC to CPU, CPU to actuators, and actuators to IOC.Each node, or device, in both configurations,must have a RM card to participate in the feedback loop. Considering the fifteen feedback loops listed in the PRD, RM can be very costly.

Point-to-point Ethernet works well as a local area network technology, but as for the LCLS rigid needs point to point may not be able to deliver. The name is a quick description of the technology involved in this network architecture. To set up point-to-point Ethernet, simply connect each device to the next using Ethernet cables. For instance, a point-to-point connection between the IOC, CPU, and actuator would connect the IOC to CPU and CPU to actuator. The concern here is that the latency would increase exponentially with the number of nodes, according to the equation for latency in this case:

15n2 + 85n + 3

Using Ethernet as the network backbone, and multicast to satisfy broad, fast communication requirements is probably the best fit. Multicast is a form of selectively communicating data to many consumer devices. On the physical layer, Ethernet cabling connects all devices, then a switch, programmed to support multicast, directs the data traffic. Further, the switch directs the traffic to a multicast group, or a set of users authorized to receive the data. Several advantages apparent in this technology is the flexibility of the configuration. If multiple CPUs need information from an IOC, multicast allows for all to receive the information, given they subscribe to the multicast group.