EMBEDDED COMMUNICAT
`
`PROTOCOL op-nous
`
`
`
`
`
`Bhargav P. Upender
`
`United Technologies Research Center
`
`East Hartford, Conn.
`
`Bhargav Upender is an assistant engineer at United Technologies Research Center.
`His research interests include protocol development, modeling and simulations,
`and performance analysis. He holds a BSEE from the University of Connecticut
`and an MEng in electrical engineering from Cornell University.
`
`Phil Koopman
`United Technologies Research Center
`
`East Hartford, Conn.
`
`Phil Koopman is a senior researcher at United Technologies Research Center. He
`currently designs and evaluates architectures and communications protocols for
`Otis elevators, Norden radars, UT automotive components, and other embedded
`applications. He has previously worked for Harris Semiconductors as an embedded
`CPU architect and the U.S. Navy as a submarine officer. Koopman holds a BSEE
`and an MEng from Rensselaer Polytechnic Institute and a PhD in computer
`engineering from Carnegie Mellon University.
`
`REMBRANDT EXHIBIT 2303
`
`469
`
`REMBRANDT EXHIBIT 2303
`
`

`
`ABSTRACT
`
`Developers are realizing that traditional low-speed, point-to-point links are inadequate for
`their increasingly complex distributed embedded applications. Consequently, they are
`investigating multiplexed communication network protocols to incorporate advanced system
`capabilities, increase reliability, and reduce wiring requirements. This paper discusses
`special considerationsfor embedded system networks, afamily tree of “standard” protocols,
`media access tradeojfs, and attractive optionsfor off-the-shelf solutions. Based on real-time
`perfonnance, cost, and hardware availability, ARCnet, CAN, and LON are strong
`contenders for most embedded systems.
`
`Embedded systems are becoming more and more complex. One of the ways to manage this
`complexity is to distribute the system functionality across several low cost microprocessors
`which communicate via a shared medium.
`
`In the past, most physically distributed embedded systems used simple point-to-point links to
`provide inter-processor communication. With increasing demand for advanced features and
`the resulting drive for more flexible and cost—effective communications, engineers are
`starting to use LAN (Local Area Network) technology in embedded systems. Most LANs
`are based on Ethernet, which is ideal for workstation-like applications having aperiodic,
`bursty communication traffic. Unfortunately, many embedded systems are unlike
`workstations in that their communication networks must efficiently support periodic traffic,
`real-time constraints, prioritized messages, and cost-sensitive applications. In this paper we
`will discusses these special considerations for real-time embedded networks, explore
`“standar ” protocols, discusses media access tradeoffs, and identify a few attractive off—the-
`shelf solutions.
`
`SPECIAL CONSIDERATIONS FOR EMBEDDED APPLICATIONS
`
`Based on our examination of several embedded applications, we believe that communication
`traffic for embedded systems tends to be mostly short, periodic messages. Because cost
`limits the network bandwidth of many applications, protocol efliciency (message bits
`delivered compared to raw network bandwidth) is very important. Efficiency is improved by
`reducing packet overhead and media access overhead. Packet overhead is all non-data bits
`added by the protocol to ensure proper routing and reliable transportation (e.g., CRC,
`address bits, acknowledgments). Media access overhead is the network bandwidth used to
`arbitrate network access among transmitting nodes (e.g., token passing). Because worst—case
`behavior is usually important, efficiency should be evaluated both for light traffic as well as
`heavy traffic. For example, Ethernet is highly efficient for light traffic but gives poor
`performance if heavily loaded. Token passing protocols have the reverse properties.
`Therefore, protocol efficiency becomes a strong metric for selecting a protocol.
`
`Due to real-time constraints of many control applications, determinacy, the ability to predict
`
`470
`
` AN DT EXHIBIT 2303
`Embedded Communication
`
`REMBRANDT EXHIBIT 2303
`
`

`
`message latency, becomes very important. Also, prioritization capability is required in some
`applications to allow quick channel access to critical messages (e.g., safety critical
`conditions) and messages in which minimum latency is crucial (e.g., sensifive control loops).
`Priorities can be either assigned to each node or to individual messages. Additionally, they
`can be either local or global. In local prioritization, each node is only aware of priorities of
`its messages, and arranges them in the transmit buffer accordingly. In global prioritization,
`the protocol allows the message or node with highest priority among all of the network to
`transmit
`
`Many applications require robust operation under extreme operating conditions. A protocol
`is robust, if it can quickly detect and recover from errors (e.g., duplicate or lost tokens).
`Some applications may require dynamic additions and deletions of nodes from the network.
`In these situations, the protocol should gracefully initialize and configure itself.
`
`Varied operating environments may dictate use of a flexible physical layer that can support
`multiple media and mixed topologies. For example, a system may require expensive fiber in
`noisy environments, but can tolerate 1ow—cost twisted pair wires in benign environments.
`Further, a bus topology may be optimum for wires, but a ring or star topology maybe needed
`for fiber.
`
`Finally, the most important consideration is the cost per node. Most of the protocols
`discussed in this paper are for high speed, high performance networks that allow expansion
`of the capabilities of a system (e.g., remote monitoring, diagnostics, and servicing).
`Therefore, the current costs may not be suitable for low-end embedded systems. However,
`with the current trend of increasing computing power and protocol support embedded in
`CPU chips, the costs are becoming more reasonable for all types of applications.
`
`PROTOCOL FAMILY TREE
`
`With the above considerations in mind, we surveyed the market for standard protocols for
`distributed applications. By identifying only standard protocols, we hoped to uncover low
`cost, ofi-the—shelf communication components and maintain interoperability with the other
`products. In particular, we hoped to discover one or two standards that were clear and
`obvious choices for embedded systems from both a technical and market perspecdve.
`
`Much to our surprise, our survey resulted in more than sixty “standard” protocols. And,
`some of these standards specifically permit the use of multiple incompatible physical
`implementations. So much for simply picking “the” standard protocol for embedded
`applications!
`
`In order to understand the relationship between these protocols, we developed a family tree
`(Fig. 1) for the most popular protocols. Most of these protocols can be well characterized as
`primarily addressing one of three different levels of standards.
`
`REMBRANDT EXHIBIT 2303
`
`Upen er/Koopman
`Embedded Communication
`
`471
`
`REMBRANDT EXHIBIT 2303
`
`

`
`-Medium Access Control (MAC): this level is part of the Data Link Layer of the
`Open Systems Interconnect (OSI) seven layer reference modell. This low-level
`sublayer defines the rules for bus sharing and arbitration. Every communication
`network uses one of these fundamental MAC protocols.
`-Protocol Implementations: this level consists of hardware/software implementations
`of a MAC scheme. Market forces have made some of these protocols, the defacto
`standards in their application areas (e.g., Ethemet, ARCnet).
`-High Level Standards: this level represents protocols that are developed by world-
`wide standards committees. These standards are trying to provide cohesion and
`interoperability by addressing the higher, application layers of the OSI model.
`
`MEDIA ACCESS CONTROL MECHANISMS
`
`In order to make sense of this tangle of standards, we will proceed from the low level to high
`level. MAC protocols determine the basic technical merits of any communication network.
`Once we understand each MAC scheme, we can then see how higher level standards fit them
`together.
`
`Connection Oriented Protocols
`
`Before LANs became popular, connection—oriented protocols were heavily used to connect
`remote terminals to mainframes. Usually, the nodes are connected using point-to-point links
`(telephone wire, serial line, etc.). Communication between two nodes requires physical
`connection using handshaldng signals, or logical connection via intermediate nodes.
`Connection
`
`based protocols are deterministic between physically connected notes, and have readily
`available hardware and software. For an embedded system with modest communication
`requirements, this might be a cost efiective protocol. Sometimes, this type of protocol is
`added to a more complex communication system to provide backward compatibility to older
`systems (e.g., BACnet1). This type of protocol is used by the X.253 public network standard
`(network services offered by telephone companies) and IBM’s System Network Architecture
`(SNA3).
`
`Polling
`Polling is one of the more popular protocols for embedded systems because of its simplicity
`and determinacy. In this protocol, a centrally assigned master periodically polls the slave
`nodes for information. Since polling is done through some type of token (special string of
`bits) passing, this protocol is also known as the Master/Slave Token Passing or MS/I‘P. The
`majority of the protocol software is stored in the master and the communication work of
`slave nodes is minimal. This protocol is ideal for a centralized data acquisition system
`where peer—to—peer communication is not required. However, fora more complex embedded
`system, the single-point-of-failure from the master node is unacceptable. Additionally, the
`
`L
`
`472
`
`Upender/Koopman
`Embedded Communication
`
`I
`
`REMBRANDT EXHIBIT 2303
`
`

`
`polling process has high MAC overhead and limited capabilities. These protocols have been
`standardized by the military (MIL-STD-1553B‘ and MIL-STD-17735) for aircraft
`_
`.
`subsystem communications. Some variants of this protocol allow inter-slave communication
`through the master and multiple masters (e.g., Profibus‘) for redundancy.
`
`ect'on
`Co
`Bin
`Countliloxvn M I
`
`C MA/CD
`CSMA/CA S
`
`T k
`0 en Bus
`
`T k Ri
`0 en
`
`ng
`
`TDM
`
`IEEE 802.3
`
`/P
`Ethernet ARCNET
`
`FND CEBus
`
`Automotive
`
`Building
`Automation
`
`B3¢kb°"°/
`Aerospace!
`Military
`
`Figure 1: “Standard” Protocol Family Tree
`
`Time Division Multi le Access TDMA
`
`TDMA is heavily used in satellite communications: but is applicable to local area networks
`as well. In this protocol, each node transmits during its uniquely owned preallocated time
`slot To maintain clock synchronization among all the nodes, a bus master broadcasts a
`frame sync signal before each round of messages. Like polling, TDMA is a simple protocol
`with deterministic response time that is well suited for balanced (evenly distributed), fixed
`length messages. Weaknesses include the bus master constituting a single-point-of-failure
`and bandwidth wasted by slots reserved for idle nodes. If a slot is not being used in some
`variations of TDMA, all stations can advance to the next slot early to conserve bandwidth
`(variable length TDMA). Time based protocols have been popular in military aviation
`applications. For example, DATAC‘, Digital Autonomous Terminal Access
`Communications, is being used by NASA and Boeing.
`
`Token Ring
`In a token ring, the nodes are connected in a ring—like structure using point—to-point links. A
`single token signal (special string of bits) is passed from one node to another around the ring.
`
`REMBRAN DT EXHIBIT“2303—'—' “‘Upe‘nae'r/1<oop‘m“an'
`Embedded Communication
`
`473
`
`REMBRANDT EXHIBIT 2303
`
`

`
`The holder of the token has access to the network. This protocol offers a bounded latency,
`high throughput during heavy traffic, and global prioritization. Under light tiafiic, token
`ring has moderate token passing overhead. Initialization of the token message, detection of
`dual tokens or token loss adds to the complexity of the protocol. Many users are concerned
`that a break in the cable or a failed node can disable the network. However, node bypass
`hardware and dual rings can be used to address this concern’. Since the ring topology is
`unidirectional, it is well suited for fiber optics. Consequently, many LANs and Wide Area
`Networks (WANs) are moving to this type of protocol. For example, FDDI (Fiber
`Distributed Data Interface) uses dual counter-rotating rings to achieve higher reliability than
`bus or star topologiesw.
`
`Token Bus
`
`The operation of token bus is very similar to the token ring — a token is passed over the bus
`from one node to another creating a virtual ring. It works well under heavy traffic with a
`high degree of predictability. However, global prioritization of messages is inefficient, and
`latency under light loads is higher than for token ring because sharing a single bus precludes
`concurrent communication among logically adjacent nodes. Unlike unidirectional token
`ring, a break in the cable or a failed node does not disable the entire network. A lengthy
`reconfiguration process, where each node identifies its neighbors, is used to maintain the
`virtual ring when nodes are added or deleted from the network. Because bus-like topologies
`are well suited for manufacturing plants, MAP”, Manufacturing Automation Protocol,
`adopted this protocol. Additionally, ARCnetu, Attached Resource Computer Network, uses
`this protocol for LAN connectivity and process control. Adaptive Networks’ PLC-192
`power line carrier chip uses a hybrid token bus protocol: under light traffic, nodes
`dynamically join and leave from the logical ring; under heavy traffic, all nodes join the ring
`to maintain stability”.
`
`Binaigy Countdown
`In binary countdown, also known as the Bit Dominance Algorithm, all nodes wait for an idle
`channel before transmitting. Competing nodes resolve contention by broadcasting a signal
`based on their unique node identification value. During this transmission, a node drops out
`of the competition if it detects a dominant signal opposite to its own; thus, if a “l” signal is
`dominant, the highest numbered transmitting node will win the competition and gains
`ownership of the channel (Figure 2). It is also possible to arbitrate using unique message ID
`values rather than the node IDs to implement globally prioritized messages. This protocol
`has good throughput under light loads. A problem is that since all messages are prioritized,
`it is difficult to achieve fair access for all nodes under heavily loaded conditions. Using this
`protocol, Bosch developed a complete Controller Area Network (CANR) specification for
`the automotive applications. The Society of Automotive Engineers standard SAE J-185015
`also uses this protocol.
`
`474
`
`UP¢"3€f/K°°Pm3“
`Embedded Communication
`
`REIVIBRAN DT EXHIBIT 2303
`
`REMBRANDT EXHIBIT 2303
`
`

`
`7
`
`Node ID
`
`MSB
`
`Node 72
`(0l00l0OO)
`Node 71
`(01000l1l)
`
`Node 42
`(00l0l0l0)
`
`
`
`Node 42 loses here
`
`Node 1 loses Ehere
`
`2 TIME srors
`
`Figure 2: Bit Dominance Procedure
`
`Carrier Sense Multiple Access with Collision Detection §CSMA[CD)
`CSMA/CD has been widely researched with a large number of published variations9'l6' In the
`simplest case, a node waits for the bus to go idle before transmitting. If multiple stations
`
`transmit simultaneously (within two bus propagation delays), the messages collide. The
`
`nodes must detect this collision, and resolve it by waiting a randomly generated time before
`
`retrying. The key advantage to this protocol is that it supports many nodes and allows the
`
`processors to enter and leave the network without requiring network initialization and
`
`configuration. Thus, for light traffic, MAC overhead is very small. However, under heavy
`traffic the MAC overhead is unbounded, leading to a protocol with poor determinacy.
`Furthermore, detecting collisions requires additional analog circuitry which translates to
`higher costs and difficult implementation in many embedded systems. The popular
`Ethernet protocol is based on CSMA/CD.
`
`A
`lli i n Av i an
`arrier en e Multi 1 Acce with
`Many researchers have developed hybrid protocols that combine the light trafiic efficiency
`of CSMA/CD with heavy traffic efficiency of token—based protocols. The resulting protocols
`are often called CSMA/CA, or collision avoidance algorithms.
`As in CSMA/CD, nodes transmit after detecting an idle channel. However, if two or more
`stations collide, a jam signal is sent on the network to notify all nodes of collision,
`synchronize clocks, and start contention slots. These time slots, typically just over two
`propagation delays long, are assigned to each of the stations. If the number of slots equals
`the number of stations, the protocol is called Reservation CSMA or RCSMA. The RCSMA
`variation works efficiently under all traffic conditions, and global priorities can be assigned
`through slot allocation”. Using rotating slots (slots that change positions after each
`transmission), fairness can be maintained and latency can be predicted. However, because of
`
`REMBRAN DT EXHIBI mne,,Koopman
`Embedded Communication
`
`475
`
`REMBRANDT EXHIBIT 2303
`
`

`
`the one-to-one relation of the node to the slot, RCSMA is not practical for a network with a
`large number of nodes. Echelon’s LON" (Local Operating Network) avoids this constraint
`by dynamically varying the number of slots (in some cases, fewer slots than stations) based
`on expected traffic and handling the case when multiple transmitters attempt to use the same
`slot.
`
`A HAND-WAVING COMPARISON
`
`In the above discussions we have summarized the major MAC protocols and noted clear
`differences. Figure 3 shows some of the common traits and the relationships between
`various MAC protocols.
`
`
`;if6keir:-5i5asea,i
`/\
`Implicit
`Explicit
`L
`M3/Tp
`
`CSMA/CD
`1EEE802.3
`
`Token Ring
`
`/
`CSMA/CA C£‘n“§{({m
`CAN
`
`MIL-STDI553b
`xi
`Token Bus
`IEEE802.-4
`4;
`
`DATAC
`L/I
`
`Figure 3: Media Access Relationships
`
`Our opinions on the characteristics of all the media access protocols are compiled in Table l.
`The important points to notice are:
`Polling, TDMA, and connection-based protocols are simple, but do not provide
`sufiicient flexibility for advancedlsystems.
`Token—based protocols are predictable, but require complex software to maintain
`robustness.
`
`CSMA/CD is a poor protocol for hard real-time systems with heavy traffic.
`The collision avoidance protocols provide the best balance for embedded system
`requirements.
`
`THE WORLD OF STANDARDS
`
`With the understanding of the MAC protocols and sample implementation standards, we can
`discuss the high level standards. Most of these standards lack specific hardware to go along
`with the published specifications. These standards have been developed by many public
`organizations and corporate alliances at industry, national, and international levels. As one
`would expect, progress is slow and consensus is minimal. To achieve consensus, some
`
` T EXHIBIT 2303
`Embedded Communication
`
`476
`
`REMBRANDT EXHIBIT 2303
`
`

`
`
`
`
`Table 1: Media Access Tradeoffs (a hand-waving approach)
`
`organizations are compromising by endorsing multiple protocols sponsored by members,
`resulting in standards and metastandards.
`
`While high-level standards may ultimately yield benefits for high-level application
`interoperability, the compromises involved in permitting multiple physical implementations
`within a standard umbrella will likely impede standardization and cost reduction of actual
`communication hardware. For example, in the building automation industry, the Intelligent
`Building Institute (IBI) standard encompasses LON, CAB (Canadian Automated Building),
`and BACnet (Building Automation and Control Network). The MAC level of BACnet, in
`turn, allows the use of ARCnet (Token Bus), Ethernet (CSMA/CD), MS/I'P, or a dial-up
`(connection oriented protocol). So, a product that conforms to the IBI standard could in fact
`use CSMA/CA, connection-oriented, polling, CSMA/CD, or token bus protocols at the
`hardware level.
`
`In Japan, a consortium known as TRON” (The Real-Time Operating System Nucleus), is
`attempting to develop open standards for all information processing systems. They have
`defined the BTRON specification for business computing, CTRON for telecommunication
`industry, ITRON for industrial applications, and MTRON for Macro networking. In
`particular, the uBTRON‘°, a specification based on the token ring,
`is aimed at networking
`real-time microprocessors in home, office, building, and automobile automation. However,
`TRON ’s development in the embedded arena is limited.
`
`In Europe, several standards have been developed: Batibus in France, Profibus (MS/TP) and
`FND (X.25) in Germany. But their effect on the American market remains to be seen.
`
`ATTRACTIVE OPTIONS
`
`There are many options that would be ideal for a particular application, and one should
`consider all reasonable options within design and business constraints. Nonetheless, we
`think that based on real-time performance, cost, and hardware availability, the following
`
`REMBRANDT EXHIBI mner/Koopman
`Embedded Communication
`
`477
`
`REMBRANDT EXHIBIT 2303
`
`

`
`three protocols are attractive.
`
`1. ARCnet: this token-bus-based protocol has transceiver chips (COM20020), a PC-
`based development system, and various communication peripherals (routers,
`repeaters, etc...) for a reasonable cost (from Standard Microsystem Corporation). In
`fact, SMC’s new COM20051 chip integrates the COM20020 with a 8051
`microcontroller.
`
`2. CAN: this binary countdown based protocol is available from Intel (82527) and a
`Signetics/Phillips collaboration (8xC592). The 8xC592 combines the CAN protocol
`with the 8051. Development systems and supporting peripherals are offered by the
`chip vendors and other third-party consultants. The costs could drop significantly if
`automotive companies decide to endorse CAN based on studies they are currently
`performing.
`3. LON: this CSMA/CA based protocol is a contender for the defacto standard in the
`control industry. LON interfaces are available for a variety of media and provide a
`large number of predefined network services in silicon. Chips are available from
`Motorola and Toshiba.
`
`In selecting one of the standards in the family tree, we recommend that you give due
`consideration to the Media Access mechanism to avoid real-time performance problems later
`in the product life cycle. The MAC protocol should provide efficient use of available
`bandwidth, a flexible priority mechanism, bounded delays for messages, and robustness to
`failures.
`
`Looking at the family tree, it is clear that a strong communication standard for embedded
`systems is not here yet. To the degree that dilfering applications have different requirements,
`a single hardware standard isn’t possible. Furthermore, it appears that higher-level standards
`incorporate multiple protocols in response to political and business considerations rather
`than technical considerations. So, choices must be made based not only on capability, but
`also market share, and a prediction of the direction of standards in the future.
`
`REFERENCES
`
`lModin', Nasser, “The ISO Reference Model Entities,” IEEE Networks Magazine, vol. 5,
`no.4, pp. 24-33, July 1991.
`
`ZASHRAE l35P, BACnet: A Data Communication Protocolfor Building Automation and
`Control Networks, August 1991.
`
`3Jordan, Larry, System Integrationfor the IBM PS/2 and PC, Brady, New York, NY, pp. 271-
`282, 1990.
`
`478 Upcnder/Koopman
`Embedded Communication
`
`REMBRANDT EXHIBIT 2303
`
`

`
`‘MIL-STD- 1553B, Military Standard Digital Time Division Command/Response Multiplex
`Data Bus, September 1986.
`
`5MIL-STD- 1773, Military Standard Fiber Optics Mechanization ofan Aircraft Internal Time
`Division Command/Response Multiplex Data Bus, October 1989.
`
`Gzebennann, C., Kaster, L., and Rake, H., “Profibus - Open On—site Communication,”
`Proceedings of the 11th Triennial World Congress of the International Federation of
`Automatic Control, pp. 143-146, August 1990.
`
`7Sta1lings, W. S., Data and Computer Communications, 3rd ed., Macmillan, New York,
`1991.
`
`8Nationa1 Semiconductor Corporation, “ARINC 629 Communication Intergrated Circuit -
`DATAC Data Terminal IC,” ver. 2.0, Military/Aerospace Products Group, March 1992.
`
`9Tanenbaum, A. S., Computer Networks, 2nd ed., Prentice—Hall, Englewood Cliffs, NJ, 1989.
`
`“Parker, Karen and Kaufman, Steve, “Fiber-Optic Links Hold Key to Fast LANs of the
`Future,” Electronic Design, vol. 37, no. 25, pp. 46-53, December 1989.
`
`ukizzardi, Victor A., Understanding MAP, Manufacturing Automation Protocol, First
`Edition, 1988.
`
`”Hoswe1l, Katherine S. and Thomas, George M., ARCnet Factory LAN premier,
`Contemporary Control Systems Inc., Second Printing, Downers Grove, Illinois, 1988.
`
`”Gershon, R., Propp, D., and Propp, M., “A Token Passing Network for Powerline
`Communications,” IEEE Transactions on Consumer Electronics, vol. 37, iss. 2, pp. 129-
`134, May 1991.
`
`“Bosch, CAN Specification, ver. 2.0, Robert Bosch GmbH, Stuttgart, 1991.
`
`ISSAE J1850, Class B Data Communication Network Interface, July 1990.
`
`16Bertsekas, D., and Gallager, R., Data Networks, Prentice-Hall, Englewood Cliffs, NJ, 1987.
`
`“Chen and Li, “Reservation CSMA/CD: A Multiple Access Protocol for LANs,” IEEE
`Journal on Selected Areas in Communications, February 1989.
`
`RE M B RAN DT EXH I B UmnT‘*er/Kmpmao
`Embedded Communication
`
`479
`
`REMBRANDT EXHIBIT 2303
`
`

`
`m”Enhanced Media Access Control with Echelon ’s LonTalk Protocol,” LonWorks
`
`Engineering Bulletin, Echelon Corp., August 1991.
`
`”‘ The TRON Project 1992, Tron Association, June 1992.
`
`2°Tanaka, K., Yasuylci, S., Tamai, K., Tsunoda, S., and Kato, H., “Performance Evaluation of
`the uBTRON Bus,” Proceedings of the Ninth TRON project Symposium, pp. 40-45, 1992.
`
`Upen er/Koopman
`
`480 Embfddcdcommumion REMBRANDT EXHIBIT 2303
`
`REMBRANDT EXHIBIT 2303
`
`

`
`
`
`Proceedings
`
`of the Fiftb Annual
`
`Embedded Systems
`Conference
`
`V0 lu m e 2
`
`Sponsored by Miller Freeman, Inc. and Embedded Systems Programming
`
`ISBN # 0-87930-306-9
`
`©Copyright 1993 by Miller Freeman, Inc., and as designated. All rights reserved. No part of this book may
`be reproduced or copied in any manner whatsoever without the express written consent ofMiller Freeman, Inc.
`
`Miller Freeman, Inc. 0 600 Harrison Street 0 San Francisco, CA 94107 0 (415) 905-2200
`REMBRANDT EXHIBIT 2303
`
`REMBRANDT EXHIBIT 2303