X3T9.2/88-129 Note from John Lohmeyer: This file contains the minutes from both the August meeting and the September meeting. September 22, 1988 TO: X3T9.3 Fiber Optic Study Group Members FROM: Roger Cummings SUBJECT: FIBER OPTIC STUDY GROUP MINUTES Please find attached a draft of the minutes of the Fiber Optic Study Group that met at the Sunnyvale Hilton on August 12 and 13, 1988. Note that there are also nine Attachments to the minutes that relate to presentations at the meeting. Apologies for the delay in producing these minutes. This was due to personal vacation and workload, and to delays in obtaining material for the attachments. If there are any corrections required or omissions noted I can be reached as follows: Phone: Business (416) 826-8640 x3332 Home (416) 625-4074 (ans machine) Telex/MCI Mail: 650-289-5060 Fax: (416) 826-8640 x3476 Regards Roger Cummings Principal Engineer, I/O and Peripherals Systems and Strategies Group Control Data Canada Ltd. #vt/rc FIBER OPTIC WORKING GROUP MINUTES The second meeting of the ANSI Fiber Optic Working Group was hosted by Peter Dougherty of Unisys at the Sunnyvale Hilton on August 11 and 12, 1988. The group met under the auspices of the X3T9.3 committee, and an SD3 that had been approved at the Milpitas plenary. A total of 43 people attended, as follows: AMD Jim Kubinec Paul Scott AMP Charles Brill Jim Kevern Steve Yingst AT&T Kaushik Akkapeddi Ming-lai Kao Reinhard H Knerr James E Morris CDC CANADA Roger Cummings CIPRICO Stu Reile Bill Winterstein DIGITAL EQUIPMENT Tim Martin DUPONT Mike Kaplit ENDL I Dal Allan FUJITSU AMERICA Bob Driscal Jim Luttrull GAZELLE MICROCIRCUITS Chris Popat HEWLETT PACKARD Randy Haagens Del Hanson Steve Joiner IBM Peter Legakis Ron Soderstrom Horst L Truestedt INTERNATIONAL COMPUTERS LIMITED Tony Salthouse LASERCOM Glen Griffith Stewart Sando LOCKHEED Brian Peebles LOS ALAMOS NATIONAL LAB Don Tolmie MMD Paul Ralston NCR Jack Brown PARALAN/CDR Marc Brooks PCO Bill Philipson PROTOCOL ENGINES Greg Chesson Steve Cooper Larry Green SIGNETICS Nabil Damouny SUN MICROSYSTEMS Sunil Joshi TANDEM COMPUTERS Armando Pauker Phil Sinykin Jim Smith UNISYS Robert L Bergey Peter Dougherty August 11&12 Fiber Optic Working Group Minutes Page 2 The agenda had been defined by Dal Allan of ENDL Consulting to reflect the current goal of the Working Group, which is the gathering of information on fiber-optic components and existing fiber-based interfaces. To this end, presentations had been solicited from companies with existing fiber-optic interfaces as well as fiber-optic component manufacturers. A copy of the meeting notice, which includes the agenda, is Attachment 1. Dal then opened the meeting by asking the attenders to state their affiliation and reason for attending. With the exception of the solicited presenters the reasons given tended to fall into four groups, namely: a) Fiber-optic material, component, and interface manufacturers interested in applying those items for use in an ANSI standard. b) Semiconductor manufacturers researching the requirements for future, high-performance silicon or GaAs devices. c) Companies with existing SCSI or IPI interfaces looking to increase the operating distance of those interfaces. d) Companies looking for a interface for a special application where the characteristics of fiber-optics are desirable e.g. low radiation in a military environment. Dal noted that there would be no presentation by National Semiconductor on the National Advanced Systems Fiber Block Mux Channel, and that there would be no presentation of SONET due to a meeting conflict. However presentations by Ron Soderstrom of IBM on Fiber Optic Components and Testing and Marc Brooks of CDR/Paralan on a SCSI Fiber Optic Extender were added to the agenda. August 11&12 Fiber Optic Working Group Minutes Page 3 Larry Green of Protocol Engines then gave a brief description of the Protocol Engines project by way of introducing the presentation of Greg Chesson. Protocol Engines was described as a spinoff of Silicon Graphics Inc. whose objective was to produce a chipset to execute levels 3 and 4 of the OSI model unified into a "transfer layer". This concept was described as evolving quickly with a software emulation presently operational and VLSI to follow shortly. Greg Chesson then began by defining the scenario of a number of networks of different types interconnected by gateways. With prresent technology these gateways are implemented as store and forward nodes, and a considerable delay is imposed in each transit of the gateway because of the need to execute many lines software to determine routing etc. In a complex topology therefore, the performance of a communication path is therefore determined more by the performance of the gateways than of the networks themselves. Therefore Protocol Engines is working to define a chipset that can be used to create a real-time gateway i.e. one where the incoming data packet is decoded within the reception time of the packet and transmitted from the gateway with minimal delay. In order to do this it is necessary to define a new protocol which imposes a minimum of overhead, which is simple enough to be able to be handled by state machines and firmware rather than software, but which nevertheless provides all of the required features to support multicast, flow control, partial retransmission etc. A major part of the Protocol Engines project is involved in the definition of such a protocol, which has been named the Xpress Transfer Protocol (XTP). Greg then briefly described the frames of the XTP protocol before moving on to consider the VLSI implementation. Even with XTP the real-time goal was stated to be impossible with a single state machine, so the hardware design became an exercise in partitioning the processing among multiple state machines. He then described block diagrams of four different implementations, starting with the ideal and ending with that possible with current technology. Greg noted that the first VLSI implementation is expected to operate at speeds compatible with FDDI. Unfortunately the slides used by Greg in his presentation were not available for inclusion in the minutes. However Attachment 2 is a paper on XTP and the VLSI that was presented at a conference in Europe in July and Attachment 3 consists of articles on the subject of XTP that are excerpted from the Protocol Engines Transfer newsletter. August 11&12 Fiber Optic Working Group Minutes Page 4 Chuck Brill of AMP then introduced Jim Kevern and Jim presented an introduction to fiber-optics technology and components. Unfortunately Jim's slides were not available for inclusion in the minutes (some of them were unsuitable by their photographic nature), but the following subjects were covered: a) Refraction, including the definition of numerical aperture, a description of scattering and comparison of silica and plastic fibers. b) Dispersion, including modal effects and a definition of graded index fibers. c) Single mode fiber operation. d) Link Design limits, both power limits and dispersion limits and their interrelation. e) Connector types and reflection problems. f) Tapped fibers and passive stars g) Cable plant requirements Jim concluded with a note that the committee will have to develop reference models to ensure the interoperability of different component sets in the same way as has been done by FDDI. August 11&12 Fiber Optic Working Group Minutes Page 5 On Monday afternoon Rom Soderstrom of IBM Rochester gave a two part presentation. The first part contained a tutorial on fiber-optics, and the presentation was abbreviated in the light of the previous AMP presentation. The full slide set is however contained in Attachment 4 along with the slides for the second part of Ron's presentation, which concerned the use of CD-ROM type lasers in communication systems. Ron noted that the future trend of optical communications systems is towards low cost computer interconnection, and in the delivery of communications to the home - which by definition requires a very low cost link. Therefore if the considerable technical advantages of lasers over leds are to be realized in such systems the cost of laser parts has to be substantially reduced from today's levels. One way to do this is to make use of a part that is already in mass production, such as the CD-ROM laser, and to package it in a simple low cost receptacle. Because the reliability of such a component is clearly a concern IBM had established a comprehensive life test program for the CD-ROM lasers, and Ron was able to present the results of the program for components from 1976 until the present. These results are also contained in Attachment 4. Ron was closely questioned on the test results by a number of the attenders. Jim Morris of AT&T asked if optical properties of the lasers such as center frequency and spectral width were measured by the test jig. Ron replied that indivdual lasers were removed a characterized on a different jig and that no significant changes were found. Del Hanson of HP questioned the reliability distribution, and Ron stated that a log normal distribution had been observed and that the median lifetime at 50 degrees Celsius is 1 million hours. Tony Salthouse of ICL questioned the effect of temperature cycling. Ron replied that the main test was conducted at constant temperature, but that temperature cycling had been separately performed and had been found to have no significant effect on reliability. Del Hanson asked if the increase in drive current was monitored and Ron replied that it was monitored to allow failure prediction and that end of life was defined as a 30% increase in drive current. Del also asked for the highest frequency possible with the laser and the FC package, and Ron replied that the maximum could be as high as 700 MBits/s with special testing but 200-400 Mbits/s was more realistic. Don Tolmie of Los Alamos National Labs enquired if the FC package that Ron was referencing was standardized, and was told that it was a standard in Japan. There was some resulting discussion about the merits of the metal FC and new plastic SC packages (samples of both had been passed around), and the preference seemed to be for the plastic SC despite its higher cost (in the order of $10/receptacle). August 11&12 Fiber Optic Working Group Minutes Page 6 Marc Brooks of CDR/Paralan then briefly described their SF40 extender product which allows two SCSI clusters to be interconnected over up to 1000 feet of fiber. A datasheet of the extender is Attachment 5. A cost of approx $820 per end of the fiber extender was quoted. Marc was questioned on the subject of the method of handling the bus arbitration by a number of the attenders, but he was unable to provide any details. Del Hanson of Hewlett-Packard then presented the work of the IEEE 802.6 committee in defining a Metropolitan Area network. Del stated that a consensus had been reached at a November 1987 meeting to adopt the distributed queue dual bus scheme first proposed by Telecom Australia as the physical layer transport for its MAN, and he went on to describe the operation of the scheme in detail. Del's slides, and a Data Communications article on the 802.6 work are Attachment 6. August 11&12 Fiber Optic Working Group Minutes Page 7 The second day began with a presentation by Kaushik Akkapeddi of AT&T Bell Labs on AT&Ts Parallel Fiber Optic Connectors. Kaushik's slides are Attachment 7. The MAC connector is a circuit pak to backplane connector which is available in both 12 and 18 fiber versions. The plug for this connector can terminate both individual fiber and ribbon cables. A version is available as part of the Fastech standardized packaging system electrical connector. The termination of the fibers into the MAC uses a silicon v-groove chip and a set of springs which provide both transverse alignment force and an axial force to keep the fibers butted. The connectors have been subjected to a standard NTT test of 1000 insertions without damage. It was noted, however, that at present the connectors are not field-installable. The characteristics of the parallel fiber ribbon were also briefly discussed. Six and twelve fiber ribbons were described and a rough cost of $1.20/foot without sheathing was stated. Kaushik was questioned if the skew figures were effected by twisting, and he answered in the negative. He also stated that rough cost for a mated pair of connectors with fiber installed would be $25 and that in the order of a thousand connectors had been shipped as part of two AT&T products, one being a video switch. Tony Salthouse of ICL then described the Macrolan product that ICL have been shipping since 1984. Tony's slides are Attachment 8. Tony described the background to the Macrolan project, and cautioned that the amount of effort required to implement a fully layered system as per OSI is very large. He illustrated the evolution of the Macrolan concept from a token ring to simple active port switch concept that was finally adopted. Tony stated that up two three port switches could be cascaded, and described the link protocol used. He noted that one of the major benefits for the user of a Macrolan system is the freedom to physically configure a system, and that ICL had experienced a very positive response to this feature. Tony noted that there were some ICL patents that covered aspects of the Macrolan, and a list of these is also included in the attachment. It was stated that the cost of a Macrolan connection is equivalent to that of a block mux connection. August 11&12 Fiber Optic Working Group Minutes Page 8 Randy Huggins of Hewlett-Packard then presented the HP-FL Fiber Optic Link. A copy of Randy's slides is Attachment 9. The HP-FL is used to increase the distance that a disk cabinet can be placed from an HP9000 series mainframe. The HP-FL can operate over distances up to 500 meters, and provides a data throughput of 5 Megabytes/s. The fiber optic components operate at 820 nm and were acknowledged to be older types than those used in FDDI. The interfaces include buffering, and were designed to resemble the earlier and slower HP-IB (IEEE488) interfaces to simplify the creation of the driver software. The HP-FL was stated to use a proprietary 5B/6B code and some details of the link layer protocol were provided. Randy also illustrated a possible future use of the HP-FL technology as a remote SCSI adapter. He emphasized that this was not a network or a SCSI extension, but merely a remote connection. The presentations having been completed, there was a brief discussion on the functional requirements. There was general agreement on the limits of 32 stations and 256 addreesses, but much discussion on possible topologies with rings and simple point to point systems all being considered. Dal Allan concluded the proceedings by thanking Peter Dougherty of Unisys for hosting the meeting, and by selecting September 26&27 as the dates of the next meeting, with the host to be determined. He stated that the primary goal of this meeting would be to begin the definition of the functional requirements. Action items were assigned to both Jim Morris of AT&T and Randy Huggins of HP regarding packet switches. Jim was asked to arrange an introductory presentation on Fast Packet Switches, and Randy to arrange for a broad introduction to switches and the definition of a glossary of switching terms. An action item regarding a SONET presentation was noted as being outstanding from the last meeting. September 30, 1988 TO: X3T9.3 Fiber Optic Study Group Members FROM: Roger Cummings SUBJECT: FIBER OPTIC CHANNEL WORKING GROUP MINUTES Please find attached a draft of the minutes of the ANSI X3T9.3 Fiber Optic Channel Working Group that was hosted by Del Hanson at Hewlett Packard, San Jose on September 26 and 27, 1988. Note that there are also seven Attachments to the minutes that relate to presentations at the meeting. The mailing list for this group has now grown to 85 people, and thus sending out a mailing has become very expensive, especially if a copy of all the material presented at the meeting is included. Rather than delete these attachments, it has been agreed that with the meeting notice and mailing for the next working group we will try to send only one copy of all documents to each company (or each separate division for the very large companies). It will then be the responsibility of that person to circulate the documents he receives within his organization to the other persons on the mailing list. A copy of the latest mailing list follows the Minutes and Attachments. The person to whom the mailing will be sent is designated at the "Primary" on this list. The attenders at the meeting were asked to nominate the primary contact for their organization, and I have attempted to identify the others. If there are any problems with my nominations please contact me ASAP. Note that a call was issued to all attenders at the meeting for presentations describing the following subjects: a) The OFFICE environment in absolute terms. b) Standards and local codes for running fiber cabling. These items are regarded as key to the development of a standard for the Fiber Channel. A set of Functional Requirements for the Fiber Channel was also defined at the meeting. This can be found at the end of the minutes. The date of the next meeting of the ANSI X3T9.3 Fiber Optic Channel Working Group has been set at December 1&2, 1988. It will be hosted Armando Pauker of Tandem Computer in Cupertino, CA. If there are any corrections required to, or omissions noted from, the minutes I can be reached as follows: Phone: Business (416) 826-8640 x3332 Home (416) 625-4074 (ans machine) Telex/MCI Mail: 650-289-5060 Fax: (416) 826-8640 x3476 Regards Roger Cummings Principal Engineer, I/O and Peripherals Systems and Strategies Group Control Data Canada Ltd. #vw/rc MINUTES OF THE THIRD FIBER OPTIC WORKING GROUP MEETING The third meeting of the ANSI X3T9.3 Fiber Optic Working Group was hosted by Del Hanson at the Hewlett-Packard Optical Component Division facility at 350 W Trimble Road, San Jose, CA on September 26&27, 1988. The meeting notice was circulated by Peter Dougherty of Unisys. A total of 33 people attended, as follows: AMD Neil Mammen AMP Charles Brill ARDENT Dyke Shaffer AT&T Ming-lai Kao James E Morris AT&T BELL LABS Mark Pashan AVANTEK Gary LaBelle CANSTAR Karl Lue Shing Kumar Malavalli CDC Wayne Sanderson CDC CANADA Roger Cummings ENDL I Dal Allan FUJITSU AMERICA Bob Driscal Jim Luttrull HEWLETT PACKARD Randy Haagens Del Hanson Steve Joiner IBM Earl Johns Horst L Truestedt IBM GENERAL PRODUCTS DIVISION Curtis Wong LASERCOM Stewart Sando NATIONAL SEMICONDUCTOR Paul Sweazey OPTIVISION Antonio Dias Mike Hauke Anil Jain Edward Onstead TANDEM COMPUTERS Armando Pauker Duc Pham Phil Sinykin Jim Smith UNISYS Gibson Carter Peter Dougherty David Mortensen September 26&27 Fiber Optic Working Group Minutes Page 2 The meeting was opened by the Chairman of the Working Group, Dal Allan of ENDL Consulting, who distributed an agenda for the meeting. The agenda is Attachment 1. Dal noted that the presentations by Protocol Engines on details of the XTP protocol and Scientific Computer Systems had been postponed to the next working group meeting, and that the overview of SONET had been arranged to take place during the Boston plenary meeting. He also stated that no-one had yet agreed to present on the subject of Futurebus. The presentations then began with a review of High Performance Switching Technologies given by Mark Pashan of AT&T Bell Labs. A copy of Mark's slides is Attachment 2. Mark began by considering the evolution of packet switching, and three generations were identified, namely: a) The earliest systems in which the switching is performed in computer memory. These systems were said to be capable of the order of 1000 packets per second. b) Second generation systems in which the switching is performed by Time Division Multiplexing techniques by units interconnected by a LAN or bus and under the control of a centralized computer. These systems are capable of in the order of 100K Packets/s. c) Currently proposed systems that are true parallel processing systems in which switching elements are interconnected into Banyan, Richards, or other types of network. Mark then moved on to consider the ways in which the switching fabric could be enhanced, and to detailed consideration of Richards networks, which had been implemented with ECL logic in the main switcher of the AT&T VIVID video conferencing scheme in the early 80s. One of the advantages of the Richards network was stated to be the simple algorithm for making additional connections. Mark was asked if the switch carried analog or digital traffic, and he replied that once the switch had made a connection then it could carry either sort. Mark then moved on to consider the "Knockout" principle in which packets pairs undergo a number of "contests" to determine their relative priority. The advantages of this scheme were identified as minimizing both buffer requirements and throughput delay. He stated that because this is still a fully-interconnected system its realization requires long, heavily loaded, bus structures to operate at very high speeds, and presented a modular architecture and an interface design to address such problems. September 26&27 Fiber Optic Working Group Minutes Page 3 A Wideband Packet Switch based on a Banyan network was considered next, and network operation was illustrated in association with Trunk Controllers that provide address translation and queuing. All of the 2x2 switching nodes in the network were described as being identical. Two routing examples were given and it was noted that there is still only a single path between any input and output in the network, and thus contention at any node in that path causes the entire path to be blocked. Mark gave some details of a field experiment that was carried out in 1986 in which both voice and data were encoded in a common packet format and routed by a WPS system as described above. In this experiment a maximum packet size of 8K bytes was established by the size of the buffer at each node. The experiment network used T1 links as well as fiber optic ones, and these T1 links 120 voice connections as opposed to 24 connections of the same quality in a more usual system. Some of this compression was due to the use of adaptive pulse code modulation, and by not digitizing silence. The "Starlite" principle of non-blocking networks was then expounded, but it was noted that this does still not solve the problem of multiple packets addressed to the same output. An architecture which overcame this problem with efficient use of buffering was however described, and this was followed by two examples of the operation of Batcher sorting networks. Mark then concluded his presentation with a description of the evolution of Electro-optical and pure Optical switching fabric technologies. The first two phases were described as consisting of photonic interfaces to an electronic fabric, and a lithium niobate fabric controlled by electronic means. Mark also anticipated a future multichannel, multihop system where the switch architecture would be based on the knockout principle and an optical bus (multiple signals wavelength multiplexed onto a single fiber). Del Hanson of Hewlett-Packard remarked in passing at the end of the presentation that an agreement has been made for broadband ISDN and the 802.6 MAN to use an identical packet header, and thus the 802.6 MAN could see use as an access network. September 26&27 Fiber Optic Working Group Minutes Page 4 Randy Haagens of Hewlett-Packard then presented an introduction to switching types and terminology. A copy of Randy's slides is Attachment 3. Randy began with a statement of what he saw as the problem facing the group, that of the definition of a subnet capable of connecting multiple SCSI and IPI busses to multiple processors and supporting the direct connection of devices supporting the IPI and SCSI command sets. He then presented alternative topologies for such a subnet. This viewpoint caused the subject of the differentiation of a channel and a network, and the possible utility of FDDI, to be thoroughly revisited by the committee. The following differentiations were agreed between a channel and a network: a) a channel has a known environment and topology b) a channel has limited addressability c) in a channel the path is implied by the address It was agreed that FDDI could and should be used as the subnet to connect multiple processors to multiple IPI and SCSI clusters. But the goal of the working group was reaffirmed as the development of a faster and extended distance channel with a physical protocol overhead and latency equivalent in percentage terms to today's channels. The protocol complexity, access delay and retiming required by FDDI's general networking approach make it impractical to use it in this application, although it was again stated that the FDDI multimode PMD would merit strong consideration for use in the Fiber Channel. Randy then enumerated five types of switching, and reviewed a matrix detailing the features of each. He stated that no practical fast circuit switch had yet been implemented. Roger Cummings of Control Data Canada asked Randy how he would classify the combination of the Xpress Transfer Protocol and the Protocol Engine VLSI. Randy replied that it was an system that implemented more functions in hardware to raise speed, but which otherwise differed little from a traditional approach. September 26&27 Fiber Optic Working Group Minutes Page 5 Dal Allan then described the current status of the SD3 for the Fiber Channel effort. Dal stated that it had been passed by X3T9.3 to X3T9, had passed the X3T9 letter ballot, and had been passed to X3 where it had again been sent for letter ballot with a 36-1 result. The negative vote was returned by AMP, and Chuck Brill of AMP elucidated the reasons behind it. These were to request the use of the expertise of other committees in the area of fiber optic connectors, and to not to initiate another definition as was implied by the SD3. A copy of the AMP statement that accompanied the negative vote is Attachment 4. Dal Allan clearly expressed the view of the meeting when he stated that the working group would seriously consider ANY means of avoiding a repetition of the extended connector debate that had been experienced in X3T9.5. Dal Allan then lead an attempt to further refine the requirements for the Fiber Channel. This was based upon the requirements agreed at the First Working Group meeting, which Dal had for convenience summarized on a single sheet which is Attachment 5. This effort occupied the rest of the first day of the meeting, and nearly all of the second day. For convenience the final set of agreed requirements is placed at the end of the minutes. Dal began the definition with a description of the present IPI Physical protocol, which he stated incurred 5 turnaround delays for each burst of data transferred. Clearly this is unacceptable in a long channel implementation, and thus a much different type of physical protocol is required. This combined with the serial nature of the interface causes a scheme in which the physical protocol information is added to the front of the higher level's command and data packets to be considered. The Xpress Transfer Protocol has demonstrated that an overhead of 1.5 turnaround delays per burst is possible with such an approach. September 26&27 Fiber Optic Working Group Minutes Page 6 Dal was questioned on the need for a channel length in the order of hundreds of meters, especially in the light of the present SCSI length restrictions. He responded by describing a complex site in which SCSI devices were connected to multiple high performance hosts via an IPI-SCSI controller. The rationale for this was clear - the device bandwidth requirements were low enough that SCSI was the only interface required and therefore available, but there was still a need to connect such a device to a high performance host. Thus the physical topology of the site would be much more flexible, and the architecture much simplified, if the device could be directly connected to the remote host over a Fiber Channel that supported the SCSI protocol and not though the IPI-SCSI controller. Dal stated that these complex sites are a prime target for the Fiber Channel, and that the ability to "unify" the IPI and SCSI command sets on a single physical interface is a major advantage to such sites. Also the existence of a parallel copper version of the fiber channel would allow the shorter length interfaces in such a configuration to be software compatible with the fiber ones without incurring the cost of the fiber components. Wayne Sanderson of Control Data suggested that a serial copper version would also have some merit for short distance applications, and that it would be as simple as bypassing the fiber optic transmitters and receivers. Jim Morris of AT&T and others cautioned that this may be too simplistic due to the dependency of things such as the coding and clock recovery schemes on the characteristics of the media and the fiber-optic transceivers. It was generally agreed, however, that a serial copper version should be included. Dal's complex site also emphasized the use of the Fiber Channel with multi-ported devices, and this lead to the ability to diagnose failed channels on-line being added to the fault tolerance group of requirements. September 26&27 Fiber Optic Working Group Minutes Page 7 The second day of the meeting began with work continuing on the amplification of the requirements. However this was suspended at one point to allow two presentations to be made by Optivision. The first of these was given by Anil Jain, who gave an overview of a number of existing products and ongoing projects. Unfortunately Anil's slides were not available for inclusion in the minutes. The products included optical interconnections, a crossbar switch with optical interfaces that is being considered by DARPA for use as an "optical backplane", and bus extenders that use fiber optic links. The project that generated the most interest was that to create an all-optical crossbar switch. This is a non-blocking switch which uses an electrically controlled shutter to regulate the passage of light. Incoming fibers are split and connected to multiple shutters, for example in one column of the shutter matrix. Outgoing filters would then be split and connected to a row of the shutter matrix. Optivision have built a 4x4 prototype of the all-optical switch, and quoted an overall loss figure of 19 dB per connection which included the input and output splitting loss. Del Hanson asked if this term had the usual switch interpretation and Anil replied that it merely identified that there were four sources of information and four destinations. Optivision is also close to completing a 16x16 switch, and Anil estimated the technology used is twice as compact as that in the 4x4 which allows the entire switch to be contained in a nineteen inch rack that is 10 inches tall and 24 inches deep. He stated that multimode fiber is used, and the current design uses 50 um and 200 um fiber for input and output respectively. The individual channel rate is one gigabit. Mike Hauke then took over to describe Optivision's bus extender product family. A copy of Mike's slides is Attachment 6. Mike began by considering possible design approaches for a VME bus extender, beginning with a "direct mux" approach in which all signals are directly sampled, their states encoded and transmitted over the connecting fiber. This approach is simple, but it requires a very high bandwidth to be supported by the connection, and this implies a high cost. An alternative approach is to provide a memory to memory channel, and although this approach is not software transparent it is much more cost effective. Mike noted that bus extenders are required to have the same bit error rate as a backplane rather than a comms channel, and quoted a rate of 10**-15 for the memory to memory extender. Del Hanson asked how this rate had been verified, and Mike replied that it had been justified by extrapolation assuming Gaussian characteristics, and agreed that it was therefore subject to some uncertainty due to the characteristics of a "real" environment. September 26&27 Fiber Optic Working Group Minutes Page 8 Mike then moved on to a detailed consideration of Optivision's fiber optic DR-11W interface extender. He began with an analysis of the DR-11W cable interface, and made it clear that the lack of a firm definition of interface usage was one of the major problems to be overcome in the extender design. He even knew of one application which merely used the data lines with no strobe, or use of any other control signal. Thus a signal sampling approach was indicated, with the sampling including prioritization of multiple simultaneous bit changes. The signal transitions are then encoded for transmission down the fiber optic link. A through delay of 1us for a single bit change was quoted. Curtis Wong of IBM asked if a CRC was appended during the encoding process. Mike replied that parity bits were appended to the data lines, and the control signal transitions were encoded to detect single bit errors. He stated that upon detection of an error the extender shut down and an interrupt is generated to the cpu, which then has responsibility for restarting the extender. At a 10**-15 bit error rate the interrupt period is predicted to be 1800 hours. Mike also described a High-Speed mode that in which the extender generates false handshakes and buffers the data in a FIFO. Roger Cummings asked for a description of the physical configuration of the extender, and Mike replied that two versions are available - a double Eurocard and a standalone cabinet. Dal Allan enquired as to the status the extender, and Mike replied that the unit was undergoing beta testing. Dal further inquired about the marketing strategy, bearing in mind the low cost nature of the DR11-W interfaces themselves. Mike stated that, despite the low interface cost, expensive equipment is often connected to the DR-11W and quoted an example of a beta site that uses the DR-11W to connect a high-performance image processing workstation to a VAX. He saw the primary appeal of the extender in allowing present implementations to be extended rather than require a re-design for to create a superior interface. Mike noted in closing that the 820 nm multimode fiber link could be easily upgraded to 1300 nm as the AT&T transmitter used has a plug-compatible 1300 nm replacement. The working group then resumed the definition of the requirements, with the results as shown at the end of the minutes. September 26&27 Fiber Optic Working Group Minutes Page 9 Dal Allan concluded the proceedings by expressing the group's thanks to Del Hanson for agreeing to host the meeting at very short notice, and for providing such excellent facilities. December 1&2 were agreed as the dates of the next meeting, and Armando Pauker kindly agreed to host the meeting at Tandem. Stewart Sando of Lasercom had asked to present to the meeting, but because of schedule conflicts this proved to be impossible. Datasheets of Lasercom's fiber optic components are however included as Attachment 7. A call was also issued at the meeting for presentations describing: a) The OFFICE environment in absolute terms. b) Standards and local codes for running fiber cabling. This is a key area that has to be addressed before formulation of the standard can begin, and thus presentations are solicited from ALL group members. September 26&27 Fiber Optic Working Group Minutes Page 10 FUNCTIONAL REQUIREMENTS/REFERENCE MODEL APPLICATION 1) Resources dedicated to a single operating system 2) Closed System (i.e. defined set of resources) 3) No provision for inter-networking 4) Known topology 5) Path implicit from address 6) Master-Slave relationship may be dynamic 7) Designed for the control of peripherals 8) IPI and SCSI devices can coexist on the same channel 9) Multiple processors can coexist on the same channel 10) Coordinated access to shared peripherals 11) Compatible with existing SCSI and IPI-3 command sets COST/DISTANCE 1) Able to use multiple technology choices 2) Achieve approximate cost parity with a copper connection in an equivalent environment. 3) Medium designed to accommodate future product generations 4) High performance and long distance operation possible at a cost premium that scales with the parameters 5) Cost effective at 50 meters, and designed to promote operation over longer distances ENVIRONMENT 1) Office compatible in terms of "weather" and shock and vibration 2) User education requirements equivalent to FDDI, and greater than for the existing copper interfaces September 26&27 Fiber Optic Working Group Minutes Page 11 FAULT TOLERANCE 1) Feasible to create configurations with no single point of failure 2) Support multiple pathing 3) Continue operation with attachments powered off 4) Support on-line servicing 5) Permit operation-critical active elements if have acceptable MTBF 6) Support system detection and isolation of all failures (included tolerated ones) SPEED 1) Repetition rate dependent on technology PERFORMANCE 1) Performance expressed as burst data rate in real bytes 2) Granularity of 5, 10, 50 and 200 Megabytes 3) 80% protocol efficiency @ 4K burst per basic cable 4) Hardware error detection on TBD boundaries 5) Flow control supported DISTANCE 1) Distance can vary with different PMDs: Optical - 5m to 2Km with a median of 500m Copper Serial - 0 to 10m ERROR RATES 1) Random bit error rates on medium are product of the technology 2) Error rate of catastrophic events (ex. control protocol failure) are 10(-12) on optical 3) Aim for copper to achieve same rate. (Have EMI concerns) September 26&27 Fiber Optic Working Group Minutes Page 12 PROTOCOL ETC. 1) Low latency control protocol (datagram) 2) Minimize station delays 3) Provide for power sequencing 4) Limit number of connections to 32 5) Channel length (of all connections) measured at 20 us (4Km) round trip for performance judgments 6) A connection is the sum of components from PC trace to PC trace 7) A connection will offer at least the same MTBF as a SCSI or IPI today 8) A limit will be set for the number of passive connector pairs between active parties 9) Connector footprint <= equivalent SCSI or IPI today 10) Fiber channel can achieve higher data rates using parallel data paths. 11) Fit confined spaces with tight cable bend. (Not worse than today's SCSI and IPI)