September 20, 1989 TO: X3T9.3 Fiber Optic Study Group Members FROM: Roger Cummings SUBJECT: FIBER CHANNEL WORKING GROUP MINUTES Please find attached a draft of the minutes of the ANSI X3T9.3 Fiber Channel Working Group of September 11 and 12, 1989. Note that there are also seventeen Attachments to the minutes that relate to presentations at the meeting. Note that this package of minutes and attachments will be directly mailed only to those persons who have attended at least one of the two most recent working group meetings. The full package of minutes and attachments will continue to be included in the regular X3T9.3 mailing that results from the plenary meeting following the working group (in this case the October mailing). Thus interested parties that are unable to attend the working group meetings are strongly advised to subscribe to that mailing. The next Fiber Channel Working Group will be held of the Wednesday of the October Plenary Week (October 18), which is being hosted by IBM at the Howard Johnson Hotel and Conference Center in Raleigh, NC. A meeting notice for the plenary week is attached along with a schedule of X3T9.3 meetings (both plenaries and working groups) At the October X3T9.3 plenary meeting, a formal vote will be proposed to accept the recommendation of the September Working Group that the IBM 8B/10B coding scheme be specified for use in the Fiber Channel. This vote will require a simple majority of X3T9.3 members to be adopted. However before this vote is proposed it is intended to propose a Standing Order by which any technical decision that is adopted by an X3T9.3 plenary and that stands for more than two months will require a two thirds majority to be overturned. This order is intended to avoid delaying the completion of projects by revisiting the same technical issue multiple times. The agenda with regards to the vote on the coding scheme will be as follows: a) On Tuesday morning it is intended to give the proposers of all coding schemes that have been presented to the Working Group as of the September 1989 meeting an opportunity to reprise their arguments if they so desire. A time limit of 30 minutes will be enforced per presentation. b) On Tuesday afternoon, the vote on the coding scheme will be called. Note that if the business of the plenary should conclude with the coding scheme vote, then the Working Group meeting will commence on Tuesday afternoon. If there are any corrections required to, or omissions noted from, the minutes I can be reached as follows: Phone: (303) 673-6357 (Business) (303) 665-0761 (Home) Telex/MCI Mail: (650) 289-5060 Fax: (303) 673-5891 Regards Roger Cummings Senior Engineer Subsystems Controller Development MD4271 #ncd0/rc MINUTES OF THE TENTH FIBER OPTIC WORKING GROUP MEETING The Tenth meeting of the ANSI X3T9.3 Fiber Optic Working Group was hosted by Terry Anderson of Ancor Communications at the Embassy Suites Hotel in Bloomington, Minnesota on September 11 and 12, 1989. A total of 48 people attended, as follows: AMD Jim Kubinec John Pottebaum Paul Scott AMDAHL Rich Taborek AMP Bob Southard Robert N. Weber ANCOR COMMUNICATION Ron Benton Bill George AT&T MICROELECTRONICS Phillip Fraley BT&D TECHNOLOGIES Ray Johnson CANSTAR Kumar Malavalli Warren Taylor CDC Lee Hartung Wayne Sanderson CODENOLL TECHNOLOGY CORP. Frederick Scholl CONVEX COMPUTER CORPORATION Gary Stager CRAY RESEARCH INC. Marvin Bausman Dennis Nessith John Renwick Wayne Roiger DIGITAL EQUIPMENT CORP. Chris Baldwin Alan Kirby Kent Springer DONAVAN INTERNATIONAL Don Pederson ENDL I Dal Allan FORD AEROSPACE Gary Waldeck FUJITSU AMERICA Bob Driscal HEWLETT PACKARD LABS. Chu Yen HONEYWELL OPTOELECTRONICS Bob Biard HONEYWELL SSPL Thomas Lane Jerry Quam IBM Joseph R. Mathis Ken Meifert Ron Soderstrom Horst L Truestedt IBM RESEARCH Albert Widmer KENDALL SQUARE RESEARCH Ed Gershenson LAWRENCE LIVERMORE NATIONAL LABS. Paul Rupert John Severyn LOS ALAMOS NATIONAL LAB Wally St. John NATIONAL SEMICONDUCTOR Sam Laymoun NETWORK SYSTEMS Ken Drewlo PCO Jim Goell STORAGE TECHNOLOGY CORP. Roger Cummings Floyd Paurus Steve Zanowick SUPERCOMPUTER SYSTEMS INC. Leonard Veil US WEST ADVANCED TECHNOLOGIES Randy Foldvik The meeting began with a review of the agenda for the meeting by Dal Allan of ENDL Consulting, the Chairman of the Working Group. A copy of Dal's agenda is Attachment 1. Dal also provided a letter from IBM clarifying the situation with respect to the patents on their 8B/10B code for inclusion in the minutes. The letter is Attachment 2. Chris Baldwin of Digital Equipment then introduced his colleague Alan Kirby to give a systems presentation on the subject of the use of Forward Error Correction. A copy of Alan's slides is Attachment 3. Alan began by noting the general purpose nature of the proposed error correction scheme, and presented graphs showing the effect on different types of link errors. Jim Kubinec of Advanced Micro Devices asked if any packet size was implied in the graphs, and was told that they were completely independent of that parameter. Wayne Roiger of Cray Research asked if a triple bit error became a possibility in longer packets, and Alan agreed that a CRC might still be required at the Packet Level because of that and other mechanisms. Bob Biard of Honeywell noted that if the noise characteristics of the mesfets used in the transceivers were gaussian then multiple bit errors would be expected. Alan replied that he believed that the dominant channel errors seen were single bit types, but that switches could create additional error types. He then stated that with the proposed scheme the data link layer protocol need not handle double bit errors as they occur so infrequently that the link could be reset to recover from them. Alan also referenced an auto-synchronization feature, and Paul Rupert of Lawrence Livermore National Labs asked how this was done. The answer was that the mechanism involves shifting successive words until the correction code indicates a legal word, and this caused some concern amongst those envisaging a switched topology. Alan noted that the scheme allows the use of the frequency of corrected errors as a measure of link quality. He also noted that the throughput degradation example did not take into account network routing uncertainty. Dal Allan asked for a quantification of the improvement due to forward error correction, and a previous graph was used to indicate between four and eight orders of magnitude. Wayne Sanderson of Control Data asked if the forward error correction scheme being proposed was associated with any particular code, and was told that the Dec patents cover more than its use with the 8B/10B code as proposed. Jim Goell of PCO suggested that the decision as to the inclusion of forward error correction should be based on an examination of the underlying physical mechanisms i.e. mode partition noise, reflection noise etc. He noted that mode partition noise varies with the fourth power of the data rate (see later), so the extra overhead of the error correction code bits could drive the link into a noise floor. It was agreed that a key question to be answered in the definition process is the limitation placed on the bit error rate (BER) by noise floor phenomena. Alan was followed by Chu Yen of Hewlett-Packard, who described a silicon bipolar chipset that is used in a proprietary point to point fiber optic link. A copy of Chu's slides is Attachment 4. The link described by Chu that did not encode data but provided for control of the long term dc offset. This was done by the use of a 16B/20B or N/N+4 code in which the first two bits in the group provide a master transition to establish frame sync and the next two bits determine the frame type. The link algorithm operates by maintaining an accumulation of dc offset in the encoder, and inverting the entire 16 bit data word when appropriate to minimize the offset. This inversion is indicated by the group type bits that precede the data word. Chu provided a diagram of the baseline wander using this scheme and stated that the worst case is +/- 0.8% versus the peak value over 100,000 coded bits. Chu noted that in the first implementation of the chip set an external discrete VCO had been used, but a recent revision has integrated the VCO as well. He described the partitioning of the transmitter functions into an Encoder chip and a 4:1 Mux chip, and stated that the major reason for this was to allow the Encoder chip to use a standard CMOS process while the 4:1 Mux chip required a 10 GHz process that was new to HP at the time. Chu then moved on the consider the problems of clock recovery. He noted that in the traditional method the extraction chain loses phase significance, and that a adjustable delay line is normally used to restore phase significance with the data. He noted that this technique is not well suited to a low cost system and therefore presented an alternative in which a phase splitter is used in concert with a 20:1 divider which filters out everything but the master transition. He noted that a key part of this technique is that the two D-type latches that operate at the link rate are on the same substrate and thus track over temperature and voltage variations. He presented a phase jitter histogram of the recovered clock for the implementation that used the onchip VCO, and noted that an equivalent histogram for the discrete version would show a sigma of 3.5 ps. He showed a diagram of the Ring Oscillator VCO, which is based around a special block in which the output follows one input or the other depending upon the value of its control voltage. He then closed by summarizing the components of the chipset, and giving active device counts of 2500 for the encoder, 2000 for the decoder, 350 for the mux and 950 for the phase-locked loop. He noted that using today's technology he would expect to be able to build a single chip transmitter/receiver that dissipated less than 2.5 Watts. Jim Kubinec then presented some classical theory with regards to the effect of dc shifts, and also a comparison of the probabilistic effects of single errors on the IBM 8B/10B code and the 4B/5B code used in the TAXI chip. A copy of Jim's slides is Attachment 5 (note that the X axis of the graph on page 1 represents the signal to noise ratio in dB). Reading from the graph, Jim noted that for a BER of 10E-10, a code with a 20 % dc shift requires an extra 2 dB of s/n ratio as compared to a balanced code. With regards to the comparison, Jim noted that the 4B/5B code has a proven history in the 200 Megabit range, and with commendable frankness stated that it has a number of disadvantages for use in the Gigabit range. Albert Widmer of IBM responded to the comparison by noting that running disparity causes errors to be detected within two bytes of their occurrence and that the "comma" characters are used as a means of fast synchronization but that other, more complex, means are available if use the comma means an unacceptable error mechanism. Alan Kirby asked Albert if running disparity was always detected within the same frame as the error, and was told that a framing character is required at the end of the frame to limit the disparity to that frame. Albert then continued by noting that good dc balance means that low frequency noise can be filtered out very well, and stated the opinion that maximal use should be taken of the features obtained by dedicating the 25% of the link bandwidth to the code overhead. Alan Kirby then reprised the presentation made by Chris Baldwin at the July Working Group meeting, which proposed the use of forward error correction along with an 8B/10B code that differs from IBM's. A copy of the presentation is Attachment 6. Jim Kubinec asked if the error correction portion of the code is independent of the encoded data portion, and was told that the dc balance is from both portions and therefore a coding scheme with good dc balance and run length properties is still required. Bob Biard asked if the reason for doubling the correction bits was to maintain dc balance, and when Alan answered in the affirmative Chu Yen asked why this was necessary if the encoding of the data itself had dc balance, and was told that otherwise a single bit link error would cause multiple bit decoded errors. In answer to a question by Wayne Sanderson of Control Data, Alan noted that Dec patents cover more uses than just with an 8B/10B code. Both Paul Rupert and Albert Widmer noted that this scheme has an efficiency of 67% (or 8B/12B). Jim Goell again questioned the assumptions underlying the justification of the forward error correction, and was told that assuming a power limited system with Gaussian noise (and therefore single bit errors) the delta to the link rate balanced the correction capability. Wally St. John of Los Alamos National Labs. then shared some experience gained with Toplinc while he was at Integrated Photonics. He said that the majority of errors that had been seen in the field where multiple bit errors related to disturbances in the link environment e.g. power noise, soldering irons switching on and off etc. Wayne Sanderson suggested that these errors would be more prevalent in an open system where the environment is not a well controlled as in a closed system such as today's IPI and SCSI. John Severyn of Lawrence Livermore Labs. stated that he had reviewed the list of references contained in the presentation, and that there is some justification for the preponderance of single bit errors. Dal Allan noted that Jim Morris of AT&T had said that forward error correction would simplify the specification of a laser transmitter. Dal also asked if there was any knowledge of why the IEEE 802.6 committee had recently decided on a non error correcting protocol for its Metropolitan Area Network after a lengthy debate, but no answers were forthcoming. Bob Biard of Honeywell Optoelectronics then made a presentation on the characteristics of fiber optic receivers. A copy of Bob's slides is Attachment 7. Bob began by noting that fiber signals are inherently unipolar in that photons are not sucked back into the transmitter when it is turned off! He then moved on to describe the receiver noise sources and their characteristics, and then described the effect of the noise and the format of the data on the receiver performance. He noted that both effects are seen as phase jitter at the comparator output with intersymbol interference resulting from the coupling capacitor and the receiver bandwidth limitations, and random jitter resulting from the noise. Bob stated that for much of the last ten years his philosophy has been to require perfect short-term dc balance to allow optimal receiver performance, but that he was now somewhat relaxing that requirement and he demonstrated the effect of a deviation of 10% (25 bits in 250). He then gave a set of typical timing parameters for 1 Gigabit link, and followed this with examples of the frequency, impulse and step responses for such a system. Chris Baldwin asked if this presentation assumed constant transmitter jitter and was told that it did because of the focus on receiver sensitivity and that a similar analysis would be required at the transmitter (which would have data dependent jitter but not intersymbol interference because it is directly coupled). Chris also asked if there was a penalty at the receiver for increasing bit rates, and was told that everything can scale but that if the transmitter technology does not scale then a broadband receiver becomes necessary and in that case sensitivity drops off quickly. Paul Rupert then volunteered the information that he had received a quote from NEC for 1 Gigabit laser diode in the $100 range, and had been told that for the next step to 1.6 Gb the additional cost was 50-75%. This information was disputed by several attenders, and Bob Biard cautioned about comparing telco and shorthaul components and prices. John Severyn then lead up to the matrix comparing the different coding schemes by reviewing the basic reasons for encoding, and the operation of Hamming codes. A copy of John's slides is Attachment 8. John had calculated the number of error correction bits for data sizes out to 32752 bits, but it was pointed out that although the code is very efficient at these large data sizes the requirement to store the large data block for correction makes for a high latency and high cost implementation. John agreed with this, but stated that he had included such as scheme in the comparison matrix to act as a reference point only. Some minor changes were then made to the comparison matrix by the meeting, and these changes are reflected in the Attachment. Discussion of the matrix was then postponed until the next day. John noted in closing that the licensing is still and issue with several of the schemes, and in response to a question Dal Allan stated that the license fee typically covers maintenance costs only with an up-front $2K required to receive the relevant documentation. Dal Allan then expressed some concern that bits, baud etc. were still being confused, and produced the following matrix to aid in clarification: DATA DATA 4B/5B&8B/10B ECC+polarity FEC+8B/10B MB/s Mb/s MBaud MBaud MBaud 25 200 250 =<250 300 100 800 1000 =<1000 1267 125 1000 1250 =<1250 1500 The monday afternoon session began with Horst Truestedt of IBM introducing his colleague Joe Mathis. A copy of Joe's presentation is Attachment 9. Joe provided a comprehensive description of a link level protocol which is capable of supporting the IPI, SCSI and HSC higher level protocols across a number of topologies such as switched, broadcast hub and string types. The protocol is based upon the concept of multiple clients and servers exchanging transactions consisting of multiple frames across two separate simplex links (one in each direction). The frame consists of specific Start and End Of Frame indications surrounding a fixed format 16 byte Link Header, a data field and a fixed format Link Trailer. The Link Header includes source and destination addresses, type identification (IPI,SCSI etc), frame control, a transaction identifier and a sequence number. These fields allow link control, recovery, and connection management to be performed in a generic manner regardless of the higher level protocol being used. The Link Trailer is associated with error control and status. Wally St. John asked if the protocol handles frames being delivered out of sequence, and was told that it was not in the mindset of the presentation but that it would be a trivial extension. Dal Allan suggested that the requirement was that such an out of sequence condition be detected, but that the reaction to that occurrence could be vendor specific. Alan Kirby asked if clients and servers shared a transaction number space, and was told that the transaction number is qualified by the source and destination addresses. Joe noted that if 8B/10B code were used, then the Start of Frame field would likely be two characters long and consist of a comma followed by an identifier. He also thought that the frame control field would include a quality of service indication similar to that defined by the HSC Data Link Layer protocol. John Renwick of Cray noted the complex switched network topology, and asked if this included routing over cross-country networks. Joe replied that the protocol should work in that case, but that additional considerations would also be involved. Wayne Sanderson suggested that a single frame size be defined and this lead to a short discussion on the optimal value of that length. It was thought that with the declining per bit cost of memory a frame size of 2 Kbytes may be workable. Dal Allan strongly felt that two frame sizes should be defined - one for data and a much shorter one to contain commands and status and avoid them having to be padded. Alan Kirby noted that the number of buffers in a system could limit the system performance, but it was generally felt that enough could be provided to avoid this situation. Paul Rupert noted that the number of buffers required will be different for each topology. Joe then concluded an excellent presentation by describing how flow control would be handled by the protocol and how transfers would be acknowledged. The next presentation was given by Frederick Scholl of Codenoll Technology on the subject of plastic fiber systems and components. An article which covers much of the same ground as the presentation is Attachment 10. Frederick began by stating that today all plastic fiber systems are under 100 meters in length, and that many use a passive star coupler that is relatively easy to implement in this technology because of the large core size and the small cladding size. He described a 1986 installation for Southwestern Bell in which a 44 storey building was wired completely with fiber optics - both for the vertical backbone and the horizontal wiring through to the workstations. Although this installation used all glass fiber he saw plastic fiber today as a viable alternative for the horizontal wiring due to its much lower cost than glass fiber (for 500 micron fiber the factor is 50%) and its superior noise immunity to traditional twisted-pairs. He projected a cost for a plastic fiber interconnect of $315, and compared this to a cost today of $155 for an Ethernet transceiver. He gave the optical bandwidth of today's plastic fiber of 60 MHz at a distance of 100 meters, and stated that today plastic fiber is at a similar state of development to that of glass fiber in the early 1970s in that experimental results are better than the theory. The theoretical limit for plastic fiber is apparently an attenuation of 3-10 dB/Km. Frederick then moved on to consider plastic fiber connectors, and he circulated several samples for illustration. The connectors were said to cost one tenth of glass fiber connectors, to have an insertion loss of 1-2 dB, and to be easy to install using epoxy. He noted that the frequency of operation of most plastic fiber systems is 660 nm (red), and that most visible sources have at 50- 100 ns too slow a risetime for use in Ethernet-type systems. He described the recent development of an Indium Gallium Phosphide led which has a 5 ns risetime. Using this, he gave some estimates for an Ethernet-type system that used an active repeater and 150 meter links. He noted that fiber with an attenuation of 50 dB/Km would be necessary to build a similar system with a passive star, and that is beyond the state of the art today. It was also noted that a red laser would allow an FDDI-type system to be built with 100 meter links. Chris Baldwin asked Frederick if the detector for use with a 1000 micron fiber was not bound to be slow because of its size, and he replied that this is overcome by using a low cost molded lens for focusing. Frederick closed by identifying the trends for the future as better 660 nm sources, fiber with attenuations in the 50 dB/Km range, and smaller and cheaper connectors. Roger Cummings of Storagetek asked what the major markets are for plastic fiber today, and was told that the digital audio market in Japan in the major driving force behind developments, and that the major vendors are all Japanese. Gary Waldeck of Ford Aerospace asked if there were any plans for operation at 1300 nm, and was told that there is no good reason for such a development because the fiber has much lower attenuation at lower frequencies. Dal Allan then introduced a proposal that the Fiber Channel effort concentrate on the definition of a single mode system on the grounds that the existing FDDI Multimode PMD be specified as a lower cost alternative. He produced some correspondence from Kevin Able of Corning indicating that the worst case FDDI fiber can achieve 234 MBaud at 1 Km. Thus it is viable for a Fiber Channel that operates at 250 MBaud over perhaps 900 meters. Dal's proposal is Attachment 11, and the information from Kevin Able is Attachment 12. Ron Soderstrom of IBM questioned why a shortwave laser was included in Dal's proposal, and Dal agreed that either a laser multimode or a led multimode solution would probably be defined but not both of them. Dal Allan also distributed a definition of the fields in the burst header that had been defined at the March Working Group meeting in San Raphael, CA. This was done in response to an action item from the August Working Group. A copy of the definition is Attachment 13. The first day of the meeting then closed with a review of available multimode and single mode connector types by Bob Weber of AMP. A copy of Bob's presentation is Attachment 14. In response to a question regarding keying, Bob noted that the ST and SC connectors have a one-way key and that FDDI has 4 way keying. Wally St. John commented that keying for security may not be appropriate as at the transmit end the connectors do not have to be completely and properly mated for transmission to be possible. Bob stated that AMP offers 2.5 mm ferrules in ceramic, stainless steel and plastic, and noted that the FSMA connector cannot accommodate a positive contact finish as it requires a gap between the fibers. Ron Soderstrom stated that the SC connector is defined for single mode usage in Japan, and that it has become the standard for Nippon Telephone and Telegraph. Bob also reviewed plastic fiber connectors, and stated that AMP sees the major usage of plastic fiber for sensor connections in copiers (apparently the mini-DNP connector was developed especially for this purpose for Ricoh). Roger Cummings then distributed two one-page documents that had been generated in response to action items from the August plenary meeting. The first concerns instructions for accessing the SCSI Bulletin Board maintained by NCR. The second gives instructions for exchanging mail between Internet and the Compuserve electronic mail system. These documents are Attachments 15 and 16 respectively. Tuesday morning began with a presentation by Kumar Malavalli of Canstar on their active hub topology. A copy of Kumar's presentation is Attachment 17. He began by describing the topology as consisting of dual rooted trees (one for upstream data and one for downstream data) formed by a central hub and perhaps a set of sub-hubs. He noted that in a normal system there is only a single path between a source and a destination, but he showed that it was also possible to create a redundant configuration with two separate paths. He described the hubs as consisting of a selection side and a broadcast side, and stated that the link between the two sides is fiber in a sub-hub and electrical in the central hub. Roger Cummings asked if the protocol was different in a sub-hub and a central hub and was told that the only difference is the setting of a dip switch and that the protocol is not data rate or packet length dependent. Ron Soderstrom asked if the broadcast side of the hub could be implemented by a single optical transmitter with a passive splitter, and Kumar replied that it could be done but that a link distance of much less than todays 2 Km would result. Paul Rupert asked if both multimode and single mode versions were available, and was told that only a multimode version was available today but that a single mode version was in development. In reply to a question, Kumar noted that retiming is performed at each node at the output of the arbitration logic. He stated that the delay through each hub is approximately 400 nanoseconds, and thus the hub incorporates a small FIFO. Kumar then moved on to consider protocol variations, including a red and blue frame scheme designed to give a deterministic upper bound to the delay through a multiple hub topology. He also described a multi-path topology which can tolerate link, sub-hub and central hub failures. Paul Rupert asked how it was determined that a hub had failed, and Kumar replied that each node detected the failure separately by monitoring the broadcast side for its own message, valid data or an idle pattern. Paul Rupert asked for some price information, and Kumar gave figures of $4K for an 8 port hub with optics, and $10K for each node interface. He agreed however, that these figures probably owed more to marketing strategies than costs. Paul Rupert gave costs for a 50 Mbit switch of $3.5K per adapter and $500 per channel, and suggested that there may not be a large cost delta between a broadcast hub system and a circuit switch. Ken Drewlo of Network Systems suggested that if a true low cost system is required then one could be built using a time-slot token protocol and a passive star. This lead to a discussion on the subject of the required bandwidth, and John Renwick caused some amusement when he suggested that the bandwidth requirements start to decrease as soon a the fiber is put in the ground. As a prelude to the vote on coding schemes, Gary Waldeck attempted to quantify some of the link parameters, and this lead to an extended discussion. On the subject of dc balance there was a considerable difference of opinion, with some people believing an balance was necessary within 10 bits, and others having no problem with an offset of 25% over up to 50 bits. The Bit Error Rate also caused some controversy with some people strongly expressing the opinion that allocating a permanent 15% of the link bandwidth to a code capable of correcting single bit errors was a bad tradeoff given that even a large number of retries per day would occupy much less bandwidth. Others noted the presence of noise floors that limit the practical BER that can be achieved, and suggested that the use of a forward correction code would allow the use of lower cost components with greater tolerances. It was calculated that at 1 Gigabit the retransmission of one frame per day is equivalent to a Bit Error rate of 10E-14. Dal Allan objected to this as in his experience high bandwidth links are not in continuous use, and thus he requested that a loading factor be included. Wayne Sanderson was of the opinion that an error rate should be developed for the entire Fiber Channel system and not just the low level link. John Renwick volunteered the information Cray's HSX channels typically experience 1% usage and no errors are normally seen in transfers of the order of 100 Megabytes. This was also the experience of a number of the attenders with the Toplinc system, which has a raw BER of 10E-12 and uses 3 fibers for a total of 450 Megabaud. As regards transition density requirements, there was also a spectrum of opinion. Albert Widmer was adamant that it is not possible to build a monolithic phase-locked loop (PLL) without external components even in gallium arsenide to handle a run length of greater than 10 bits. An Anadigics part was mentioned to refute this, but this turned out to be a multi-chip hybrid. It was also mentioned that Bell Labs has a cmos monolithic PLL, but this was thought to operate at only 200 MHz. A poll of the attenders was taken, and with only two exceptions the ability to build a monolithic PLL was thought to be very important. Gary Waldeck closed by asking a significant question, which was that is there sufficient experience with fiber for long term failure mechanisms to be known. No clear answer was given, but the feeling that the experience is sufficient was noted. Jim Goell then presented some data on noise in fiber optic systems that he had gleaned by conversations with John Renwick and an authoritative source at Bellcore. The first subject was thermal noise, and for a plot of BER versus power the slope is 3 orders of magnitude per dB. Therefore a slight increase in margin renders thermal noise insignificant. The second subject was mode partition noise. Jim drew a plot of power to achieve a 10E-9 BER versus length (and dispersion) for both a forward error correcting and a non-correcting system. Over short distances the power penalty for no correction is 1.5 dB, but as the distance increases the mode partition noise becomes significant. Jim had stated earlier that the mode partition noise increased by the fourth power of the data rate, but he had learned that this is incorrect with the correct relationship being log (1-eExE2). The effect of this is that a non -FEC system "hits a wall" beyond which operation is not possible. Jim then when on to quantify the effect of mode partition noise in a "real" system. He defined the following parameters: fb of approximately 1.25 GBaud L of 1270 - 1355 nm K of 0.6 (laser mode partition noise rate) delta lambda of 4 nm (a low cost laser) telecom grade fiber In such a system the "wall", which is taken as the 1dB penalty point, was said to be around 5 Km. Chris Baldwin immediately objected to this analysis on the grounds that the theory is optimistic, and that operation over 5-10 Km will put serious constraints on the laser parameters if forward error correction is not used. Jim Goell also volunteered to produce a graph of speed versus distance for lasers of the type being considered for the Fiber Channel. The decision on a coding scheme was then approached in two stages. First the "Clear Channel with Forward Error Correction" was evaluated against the IBM 8B/10B code. The following comparison was produced: FEATURE CLEAR W/CORRECT IBM 8B/10B Overhead 20% 20% Special Characters additional free DC Balance controlled 0 Run Length >10 <10 Coding Violations no yes A vote was then taken of the attenders on the basis of one vote per company. The result was: Clear Channel with Correction 5 votes IBM 8B/10B 10 votes The second stage involved a comparison between the IBM 8B/10B code and the "8B/10B Code with Forward Error Correction" proposed by Dec. The following comparison was produced: FEATURE IBM 8B/10B DEC 8B/10B W/CORRECT BER 10E-12 10E-17 BER 10E-9 10E-14 Restrictions Tight Loose Distance OK More Median link length 500m 500m Another vote was then taken of the attenders on the basis of one vote per company. The result was: DEC 8B/10B with Correction 5 votes IBM 8B/10B 11 votes As a result of these two votes, a recommendation will be made to the October Plenary meeting that the IBM 8B/10B coding scheme be adopted as part of the Fiber Channel definition. Dal Allan closed the meeting by thanking Ancor for hosting the meeting and providing the excellent meeting facilities. He also thanked John Severyn for an excellent job of producing the comparison matrix of the coding schemes. ACTION ITEMS 1) John Severyn of Lawrence Livermore Labs. to produce a matrix comparing the features of the FDDI 4B/5B, IBM 8B/10B, and Naked coding schemes. CLOSED 3) Paul Scott to produce a definition of Bit Error Rate. 7) Henry Brandt of IBM to respond to the comparison of the error handling characteristics of the 4B/5B and 8B/10B codes. CLOSED (by Albert Widmer at the September meeting) 8) Kevin Able of Corning to define the parameters for the media of both the Multimode classes (both 50 and 62.5) and the Single Mode class. 9) Chuck Brill of AMP to define the parameters for the media- end connectors of the Coaxial, Multimode and Single Mode classes. CLOSED by Bob Weber's presentation. 10) Bob Carter to define the parameters for the patch panels of both the Multimode classes and the Single Mode class. 11) Paul Scott to define the parameters for the media and transceivers of the Coaxial class. 12) Schelto Van Doorn of Siemens to define the parameters for the transceivers for the Multimode LED class. 13) Ron Soderstrom of IBM to define the parameters for the transceivers for the Multimode Laser class. 14) Jim Goell to define the parameters for the transceivers for the Single Mode class. CLOSED 15) Gary Labelle of Avantek to try to get a representative of Hewlett-Packard experienced in their plastic fiber system to take responsibility for defining the components for the Plastic Class. CLOSED 16) Jim Goell to produce a graph of speed versus distance for lasers of the type being considered for the Fiber Channel.