Computer Services, Inc.

PCI Reflective Memory (PCI-RMS)

Features

 

  • Two DMA Engines
  • Error Detection and Recovery
  • 1.0625 Gigabits Per Second Fiber Channel (FC-2)
  • Universal PCI-Type 5 Form Factor
  • Onboard/Offboard Memory with Spinlock Support
  • PC, Workstation, Server, and Enterprise Class System Use
  • 8 MB Onboard RAM

 

Benefits

 

  • Affordable and Easy to Install
  • Simple Programming Model
  • High Performance Real-Time Clusters
  • Sun Solaris™, Windows NT®, Linux®, and TRU64 UNIX™ Compatible

 

Overview

 

PCI-based Reflective Memory (PCI-RMS) is a high performance multiprocessor distributed memory system that sets a new standard in the Reflective Memory family of technologies.  It is a foundational innovation in high-speed deterministic system connectivity and continues in the tradition of Reflective Memory and Memory Channel® interconnects.

 

Universal Type 5 Form Factor is just the latest in the evolving technologies associated with Reflective Memory architecture, which was patented in 1983 and has become a de-facto industry standard.

 

PCI-RMS is the basis for creating an industrial strength Real-Time Cluster with up to 255 nodes in a single ring.  The cluster can be configured using several different microprocessor families, motherboards, and operating systems to form a truly open heterogeneous environment.

 

Gigabaud-per-second Fiber Channel Connection

 

High performance and low latency are the hallmark of PCI-RMS.  Using our master/master ring topology, up to 255 point-to-point flow-controlled segments can be configured to form a ring up to 7.8 km long (100 feet between nodes).

 

The advantage of using a modified Fiber Channel FC-2 layer approach is it enables a simultaneous transfer of data from every node to traverse the entire ring (multipoint-to-multipoint).  Imagine, 255 nodes able to broadcast and receive at a given instance in time without tokens or data collisions in less then 300 microseconds.

 

The Fiber Channel shielded twisted-pair cable provides the highest data transfer capability of up to 1 gigabaud-per-second with the lowest cost per connection.

 

Memory Features

 

PCI-RMS memory operation and options are extremely flexible.  This technology enables you to make use of the onboard Spinlock, PCI and DMA memory features simultaneously.  The circuit card comes with 8 MB of physical SRAM.  In addition, system memory on the node CPU (up to 256 MB) can be addressed and used in conjunction with the circuit card SRAM.  The onboard SRAM can be used as hardware assisted Spinlock Reflective Memory, providing registered return receipt data transfers for added message delivery assurance.

 

PCI-RMS features low data latency of less than 5 microseconds, as follows: 2 µsec for PCI Memory to link, 1 µsec per node, and 2 µsec for link to PCI memory.

 

Additionally, we have implemented DMA engines into the Reflective Memory design.  With two onboard DMA engines, you can configure the logic circuit to function in Memory Channel mode and relieve the CPU of cycle stealing tasks such as moving blocks of memory resident data.  The two DMA engines can move blocks of data at industry leading speeds with less than 20 µsec start-up latency.

 

Additionally, if forwarding across multiple nodes is required, an average 1 µsec latency per node is incurred.  This permits data to be written to all the nodes on the link before it is written to the source node's memory.  In this manner, you can broadcast data sets to nodes in the cluster (point-to-multipoint) and create a deterministic I/O mechanism.

 

Error Detection and Recovery

 

In addition to the high performance and flexibility of the PCI-RMS, we have included features to assure that your Real-Time Cluster is manageable and your data is reliable.  PCI-RMS implements an extensive set of error detection and software recovery techniques for enhanced data integrity.  These include Cyclic Redundancy Checks (CRC) on serial data frames, bus and parity error reporting, visual status indicators, and software accessible status registers.

 

Serial Link Error, DMA Complete and Mailbox Interrupts

 

For reliable communication between connected Reflective Memory nodes, a hardware interrupt can be generated to report serial link errors when the serial link fails or a data transfer error is encountered.

 

Mailbox interrupts are used to asynchronously notify a node upon the arrival of data at the node. Nodes can also be asynchronously notified by a DMA Complete interrupt upon the completion of a local DMA transfer.

 

Summary

 

PCI-RMS is an affordable solution to your data distribution needs.  We offer a high-quality shielded twisted-pair copper media cable at several convenient lengths.  The Universal Type 5 design enables you to pick a platform to fit your project needs with quality, performance, and affordable choices.

 

 

 

Windows NT is a registered trademark of Microsoft Corporation.  Tru64 UNIX is a trademark of Hewlett-Packard Company.  UNIX is a registered trademark in the United States and other countries licensed exclusively through X/Open Company LTD.  Reflective Memory and Memory Channel are registered trademarks and RMS and Sun Solaris are trademarks of Sun Microsystems, Inc.  Linux is a registered trademark of Linus Torvalds.  All other trademarks and registered trademarks are the property of their respective holders.

Computer Services, Inc.