MINUTES OF THE ELEVENTH FIBER OPTIC WORKING GROUP MEETING The Eleventh meeting of the ANSI X3T9.3 Fiber Optic Working Group was held on October 18 at the Howard Johnson Hotel in Raleigh, NC. The plenary week during which the meeting was held was hosted by Dan Pitt of IBM. The meeting began with Dal Allan appointing Don Tolmie to the temporary position of Chairperson for the Fiber Channel Working Group. Don Tolmie set the order for 4 presentations from 3 vendors, Codenoll, Corning and BT&D. Wayne Sanderson from CDC noted that some discussions regarding the scope of the Fiber Channel Working Group would be helpful. Dr. Bulusu Dutt from Codenoll gave a presentation on plastic fiber. His presentation is Attachment 1. He stated that at a rate 250 Megabaud, the maximum operating distance today is 25 meters, but he expects this to increase to 60 meters in 1990. Kevin Able from Corning gave a presentation reviewing parameters of fiber optic cable and made a recommendation to the working group. The attached copy of his slides (Attachment 2) show the theoretical limits of optical cables. His recommendation focused on 1300 nanometer wavelengths for 62.5 micron multimode and single mode fiber. Kevin suggested that 50 micron fiber also included because of installed base. He stated that it may be best for the Fiber Group to separate from FDDI for cable plants. Schelto van Doorn from Siemens suggested that compliance with the FDDI standard would be better for price considerations. Dave Jenkins of BT&D stated that it was possible to double the current bandwidth of FDDI. Don Knudson of Digital Equipment Corp stated that the FDDI jitter budget wouldn't allow higher bandwidth. Dick Plumb from BT&D gave a presentation on characteristics of ELED devices. His slides are Attachment 3. Dick suggested that ELEDs could be used to increase the bandwidth of FDDI systems and cable plants. David Jenkins of BT&D gave a presentation discussing cost differences between bandwidths of 1.2 GigaBaud and 1.5 GigaBaud. His slides are Attachment 4. He stated that the increased speed is not the factor which dramatically increases cost, its the tight control required over the wavelength when attempting to get extremely low BER that drives the cost higher. He stressed that mode partitioning noise would be a factor at the speeds and distances we require. It's difficult to measure or predict, and reflections will make it worse. He strongly cautioned against assuming we can have very low BERs. His slides are attached. Dick Plumb presented some slides dealing with spectral density and sources of noise the transmitter may see. These slides are Attachment 5. Don Tolmie updated the Working Group with the results of the plenary vote on the group's recommendation regarding coding and baud rate. This started a discussion about lower speeds for fiber channel. Mike Pugh gave a synopsis of the current status for people who were not at the last few meetings. John Severyn from LLNL raised the issue of where the low end is (what speed). It was noted that 250 MegaBaud had been discussed as the low end. Don Tolmie stated that if any requirements existed below GigaBaud, but above 250 MegaBaud, builders of these systems should voice the requirement. John Lohmeyer from NCR stated a high volume would exist for low end connections with prices of $8 to $10, however, with higher costs, volumes would reduce drastically. Kevin Able asked what distances might be required and John Lohmeyer pointed out that SCSI requirements were outlined in the Colorado meeting. Jim Goell from PCO stated that clear indications of the possible volumes of low end usage is required to attempt an estimation of pricing. Don Tolmie indicated that there seems to be a deadlock which needs to be broken. We can't get pricing without volume estimates, people don't want to project volumes unless they know pricing. Dave Jenkins suggested that a look at what the market is today may be helpful. John Lohmeyer volunteered 3 current implementations: a 3 inch product with a 5 Megabyte/second interface, a half size product with a 10 Megabyte/second fast protocol interface, and a 68 pin site SCSI for 16 or 32 bit implementations of a 40 MegaByte/second fast protocol interface. These would all be possible target sites for an inexpensive fiber channel. Wayne Sanderson stated that a midrange requirement for large system usage exists for IPI. The Group moved to topology discussions next. Don Tolmie was unaware of any agenda for this discussion so the meeting was open for all to comment. Several people commented that they felt the goal was a point to point link and that this issue needed to be documented and made very clear to all if that isn't the goal. A suggestion was made that we need a list of agreements that have already been made detailing the makeup of a fiber channel. Wally St.John of LANL stated that it would be beneficial to state the goals of fiber channel in a written form so that its clarity is preserved. It was suggested that Dal Allan accept an action item to list significant aspects of the fiber channel that the Group had agreed on. Chris Baldwin from Digital Equipment Corp. suggested that we need to put things down in the FC0, FC1, FC2 and FC3 documents. Don Tolmie stated that the IBM proposal was FC2 and above and that it may be a good working document. John Renwick from Cray Research, Inc. stated that much was being discussed based on the IBM proposal which didn't appear in the paper. Joseph Mathis of IBM stated that sequence numbers in his proposal provided the basis for detection of an error and that through floor discussions, their usage became more complex. Richard Taborek of Amdahl Corp. cited an application which necessitated these complications. Horst Truestedt of IBM added that IPI-3 already supports packetized services, and SCSI-3 may support packetized services in the future. Bob Biard of Honeywell stated that transmission delays made these complications necessary and that the standard doesn't need to address the issue. Horst suggested that perhaps an option for usage of acks and sequence numbers would work. Joe Mathis stated he had not intended for there to be functional options. He was proposing that there were different ways or options that could be standardized. Flow control was then discussed in detail. A point made by John Renwick was that acks can do 2 distinctively different things, flow control and data delivery verification. He stated that flow control at the FC2 level is all that is required. Doug Morrissey of Unisys pointed out that a deadlock can occur if flow control is invoked and the sending FC2 needs to get a command to the receiver. He suggested 2 separate flow control mechanisms. Rich Taborek stated that if the receiver can always accept 1 command, then these deadlocks can be broken. Don Tolmie raised the question of restructuring FC0 and FC1 to match a proposal for partitioning by Michael Pugh of Integrated Photonics, Inc. In order to refamiliarize us with the concept, Michael gave a condensed version of the proposal again. Some discussion occurred regarding how complex it would be to standardize the configurations of the "modem interface" so that silicon vendors could make compatible parts. The discussion ended with an agreement that we would discuss it at the next meeting. A discussion followed about the necessity to support string configurations. Doug Morrissey commented that defining a string would take extra definition in the standard. John Renwick stated that a simple switch requires less definition than strings or rings. Terry Anderson said that his switch will place requirements on the protocol for out of band signaling. At this time, Don Tolmie summarized some key points of general agreement in the Group - Datagram service will be sufficient. Acknowledgements will be optioned with a control bit. A special expedited control frame will be supported which provides the dual flow control necessary to break command deadlocks. The group agreed that more discussion about a revised version of the IBM proposal and a review of speeds and distances for the range below Gigabaud would be discussed at the November meeting. The motion to use the IBM proposal as the current FC2 document was made by Wayne Sanderson and seconded by Terry Anderson. The motion was passed unanimously. The meeting was then adjourned.