Unconfirmed Minutes of the HSSG meeting in Wakefield, MA 5/22-5/23/96 ===================================================================== Howard Frazier brought the meeting to order. Introductions around the room were made. The recording secretary selection was made. Rebecca Farley (too new to know any better) was elected with the promise this would be the one and only time. Copies of the minutes from the La Jolla meeting were passed out and would be approved later. Howard presented slides on: P802.3z Status Report NESCOM will review PAR on June 19 IEEE Stds Board will consider PAR on June 20 E-mail Reflector Currently ~300 names on reflector Web Site Howard requested that everyone strive to take advantage of this service. ftp://stdsbbs.ieee.org/pub/802_main/802.3/gigabit" will always contain a file called MEETING.TXT for information on future meetings. Preliminary Outline for P802.3z Will use 100BASE-T outline as a starting point Technical Presentations Modifications were suggested for the copper presentations to be moved above the optical presentations for those who have time conflicts. One additional presentations was added in the optical sections by Steve Swanson. Two presentations were not made, Andy Bechtolsheim and G. Ramikrishnan. Technical Presentations. See handouts. REPEATER / Bit Budget ===================== Mohan Kalunte - 1Gb/s CSMA/CD - Performance Simulations - 9:30-10:10 Questions were raised about the latency comparisons Bernard Daines - Gigabuffer Repeater Proposal - 10:10-11:00 Q: Why use CSMA/CD when it is invisible from the outside. R: That's not what is being presented. Q: The definition of a repeater is not a store and forward device. R: The specifications change and we may need to define something new. Break 11:00-11:15 Questions were raised about the voting rules. Geoff Thompson clarified that there are 3 plenaries a year. 75% attendance at two of the plenaries is sufficient for voting status. One may also substitute 75% attendance at a plenary with 100% attendance at an Interim meeting. Steven Haddock - Slot Time Issues - 11:25-12:20 Q: With 512 slot time min. the re-transmit size goes up. R: Yes. Q: What about late collisions. R: Needs a discussion. I've no solution, maybe drop last packet. Maybe you could change the rules and only re-send the last packet. Shimon: Would this change the service primitives to the MAC Client. R: The answer is not obvious but didn't think it made a difference. Comment: Multiple packets can be received in batch mode but should be passed to the client level as individual packets. Scale down backoff - possibly limit the number of stations on a Gig network to 128 (2**7). Q: There should be concern about changing 2**10 to 2**7. It will result in more dropped packets. R: Statist- ically those packets that have already had 10 collisions will usually result in an excessive collision and be dropped. Danny Cohen - Fast Switching - 12:20-12:45 General: MsgWay is an IETF project. Proposes we do source routing. Proposes taking 1% of the addresses for the tagging scheme. Geoff: We don't have any control over those addresses. R: Then maybe this needs to be presented at another forum too. C. Wish List is Both SR and addresses, Fine Grain Flow Control, Short inter-packet interval, Type (extended), Length (i.e. longer packets), "endianness" handling, Short minimal length (0?). Conclusion: Gig nets require fast switches. Fast effective switches need source routing. Source routing can be added to 802.3 in a backward compatible manner. Lunch 12:45-2:00 MAC / PCS ========= Bernard Daines - MAC/PHY interface - 2:00-2:15 General: Introduction and overview on where we are. We need to define the interface. "Write a standard in harmony with PAR, Criteria and Objectives." Howie Johnson - Protocol for link startup and packet delimiters - 2:15-3:00 General: Focus is on PCS layer. Modification to Slide 4. ANSI X230 FC-PH. Geoff: are you proposing to do Auto-negotiation with configuration cycles. R: I'm proposing not to use Link Pulses; instead I want higher level protocol. Slide 15: C. You may want to extend the H's beyond the range of the data into the carrier extension. Q: Would prefer to see H's to the end with RR before the I's. R: I will look into it. Q: H's are inserted when there are code violations in the data stream but what about code violations in the R's. R: Never seen that; we'll have to decide what to do about that later. Q: Where are the H's inserted in the architecture. R: Code violations come into the PHY; the PHY then generates H's to the MAC. Q: Is there a minimum number of H's on the end of a packet if the error occurs after the slot time has expired. R: 1. Q: Is it a requirement that the repeater decode these vio- lations. R: Some used to record the violations but others just pass through, I'm not ready to decide this. Slide 17: Q: What happens on Remote Fault when in LINK_OK state. R: Then goes to the top state. Q: Any special (1) test modes (2) other management commands. R: (1) No (2) Possible, not defined yet. Q: Does the state machine accommodate "hot standby" link. R: No. Richard Taborek - Control Codes for link startup & packet delimiters-3:00-4:00 General: There are a few errors on the slides (9,13,14,16), see the corrected copies on line. See Fibre Channel web sites for the 8B10B coding specifics. Slide 13: Geoff: Does running disparity equal current DC balance. R: Yes. Slide 14: C. Rich's I1/2 = Howie's I3/4 and Rich's I3/4 = Howie's I1/2. Q: Why do we need preamble. R: Howie: Because we wanted the MAC media independent and there may be a media in the future that needs a preamble. Q: Some chips will trash commas; we usually send 3 in a row. R: Yes I forgot to put it in unfinished items, this needs to be addressed. Notoyasu Yano - 1.25 Gb/s serializer/deserializer - 4:00-4:30 General: Full transceiver chip includes transmitter and receiver and runs at over 1.25 Gbaud. Slide 4: Q: Was there random data on the eye pattern diagram. R: Yes. Q: Is the receiver side running at the same time during the test. R: No. Q: So there's no crosstalk represented here. R: Correct. Slide 8: Q: Is transmit side running during test. R: No. Slide 9: Q: What is the source of the data. R: Bit generator. Q: What is the clocking scheme. R: Signal generator is generating the clock. Q: I'm trying to determine if this is best case or worst case rise and fall time data. R: Best case. Recessed for the day. Stared meeting at 9:00am. Copper PMD ========== Sailesh K. Rao - MAC-PHY interface for copper PHY - 9:00-9:20 General: Stress no 8B10B format since it is a fiber PHY interface. Should be common for copper and optical 8B10B. Highly unlikely that copper will use 8B10B encoding because of the 25% data-rate penalty. Q: How much code base are you talking about. R: 9x9 instead of 8x8, so 17 extra points (that's 81 points, he means an extra 17, meaning that he is looking at 64 data points). Slide 7: Q: Have the don't cares be 1's as not to kill the code base. R: Okay. Q: In 100Base MII there is an issue on the clocking source. R: I would like to keep the clocking the same as 100BaseT; Clocks sourced by the PHY. T. J. Shan - Report on 1Gb/s transmission on copper media - 9:20-9:45 General: CAT-5, 100meters, Full-Duplex, 4-pair, Emissions class TBD. Q: On FCC approval are you targeting Class A or Class B. R: You tell me I'd like to take the easier one. C. This isn't really PAM because it is 2 axis. It is QAM. C. Their simulations have only covered 4-pair half duplex so far. Q: Did you take into consideration crosstalk an all 4 channels since you didn't use Next Canceller. R: Yes. Optical PMD =========== Jonathan Thatcher - Optical Specification Progress - 9:45-10:05 General: Will sign up for 200m on 62.5u fiber, expects to come up with spec after working parts. Slide 5: C. Gigabit will look more like FC-0 vs. FC-AL. FC-AL is going to a new specification with new measurement points. Slide 6: Q: What is RIN. R: Return noise (Relative Intensity Noise) - relative number to the signal; Light that comes back in exciting stuff that isn't supposed to be excited causing problems. Del Hanson - IEEE 802.3 1.25 GBd MMF Link Spec Development Issues - 10:05-10:40 General: Not able to meet the current distance objectives of 500m on 62.5u fiber. Slide 9: Q: Wavelengths are different. R: I made the decision to do this and these are the results; we haven't determined the wavelength range yet. Q: What I infer from slides 9 & 10 is that the difference between 770 and 820 equals a 75m difference. R: If we argue on the Modal bandwidth scale with wavelength then yes. Q: Any other advantages with specifying the wavelength. R: Yes advantages in Modal Bandwidth and attenuation. Q: Should 802.3 recommend anything to Fibre Channel. R: The discussion following these presentations should decide that issue. Q: What's the impact of 0 dispersion. R: With 0 dispersion closer to wavelength, there is a change but only slight. Q: Is there an impact with slope. R: Yes, there is an issue, the lower the slope the better you are. Q: I've been hearing VIXEL lasers are faster than CD Lasers. The specification you're working on, does it take advantage of VIXELs. R: I'm in the VIXEL camp. I would like to move out of the wavelength range, but we need to keep range for historical reasons. Q: But that costs us. R: Yes. Break 10:40-10:55 Jonathan Thatcher - Ethernet Gigabit Optical Experiments - 10:55:11:30 General: Didn't use the serializer/deserializer in my computations. Q: What kind of components were used. R: We were using "typical" stuff; we had a bag of components, picked one that was barley making spec and one normal; we looked for ways to make it bad; we could have picked components that would have yielded better results. Q: What if you wanted to use serial copper link. R: It's not the same spec.; like Fibre Channel (FC) today rather than the 40-50% eye opening they want as little eye opening as possible. Q: Note, serial copper differs from fiber that's when FC got in trouble; They did everything for fiber and got in trouble when tried to do copper. Geoff Thompson -Fiber topologies & perf. specs in contemporary stds-11:30-11:45 General: The measurements for attenuation and modal bandwidth are made precisely at 850nm and 1300nm. C. We are trying to figure out what to tell ISO; are we happy with what we've got or should we ask ISO to wait a little so we may include changes. Q: Does 850 mean 770-860 (1st window). R: Yes. Suggestion to change the objective of 500m to 300m. Steve Swanson - Fiber Specifications for 802.3 Gigabit Ethernet - 11:45-12:05 General: Put up summary of the Bob Campbell cabling distance study from 1988. The discussions were concerning the percentage of installed base of 62.5u and 50u fiber and what are the distance standards. Discussion of optical performance objectives - 12:05-1:15 Howard leads the discussion on the accuracy of the objectives. ============================================================== Fiber distance. Objective #11a: =============================== Howard: Are these distances recommended here acceptable? Discussion ensued on the accuracy of the Campbell study from 1988 to today's link distances. Bob Campbell (in reference to the pertinence of the 1988 study today): The extremes are probably not as accurate, they probably have longer distances; I believe the less then 300m is horizontal wiring, but backbone links are still pertinent. Stan: 500m was a technical limit for 10Base5 but is not relevant today. Howie: wonders if 200m gets 95% of the market then the other 5% is not worth stretching to 300m. Gideon agrees with Howie and adds that the data on the Campbell study is eight years old and a copper study not fiber. Geoff: We need to look at the sensitivity as to our position and I don't think this study does it. Jonathan: many facilities are pulling both single-mode and multi-mode fibers so this may be a moot point. Howie Ask Ahmad to reactivate the study. Ahmad Agreed to look into that. Steve: presents roadmap and says can not get 300m with .45 at 1.25G. Del: Presented slide #9 again and says Yes, with VIXELs but still marginal at 300m. Geoff: What does it take to get over 550m with 50u? Is it achievable. Stan: Yes. Del's slide #10 is put up. Jonathan wants to know how hard or soft are the requirements. It's no problem with 850 VIXEL. Howie: It is hard to understand what you are giving up. Do we make it 10 times more expensive for a 50m gain? Stan: Cost difference is 15% TX price increase to get 40m. Geoff: Propose a standards oriented direction. We take as our goal 550m @ 50u and <100m @ 62.5u. Gideon: Ask optical experts what it takes to come out to 300m on 62.5u specification-wise not design-wise. Howie presents new slide with a fill-in table for the various options of distance, cost, benefits. This table never gets fully filled in. Stan: Is there no slope in Modal Bandwidth. Steve: No. Jonathan: We want a dirt cheap solution because anything that you do to get more meters will exponentially increase cost. Gideon: But need to recognize the installed base. Steve: Support Jonathan's comments on length vs. cost. Motion: Recommend that 802.3 clarify objective 11a to read: at least 550m on 50u 500MHz km mm fiber at least 100m on 62.5u mm fiber. Move: Howie Johnson 2nd: Bill Quackenbush. Discussion: Allan: I object to the 100m, not standard. Del: Should base it on same source if you push 50u length then should push the 62.5u length. Friendly amendment: Added "with a single xcvr & link spec" for it to read: Recommend that 802.3 clarify objective 11a to read: at least 550m on 50u 500MHz km mm fiber at least 100m on 62.5u mm fiber with a single xcvr & link spec. Motion: Postpone consideration of Johnson/Quackenbush (Motion) to July. Move: Gideon Pratt. 2nd: Bob Campbell. This is a procedural (50%) vote. Yes: 35, No: 13, Abstain: 10 The Motion passes. Straw poll on how many people think 550m @ 50u and 200m @ 62.5u would be adequate. Yes: 32, No: 3. Collision. Objective #12/Slot Time: =================================== C: Slot time effects collision domain - There's no objective for this and IPG. BER differences - no objective. Pat: wants to reduce the BER from -12 to -10 and wants to know what the optic guys think. Del: It buys you very little. Allan: What about the collision domain. Howard: We said we would wait until we have a task force. Straw Poll: Collision Domains: Voting count (1) CSMA/CD with carrier extension for horizontal. 8 (2) CSMA/CD with carrier extension for limited topology (10-50) 9 (3) CSMA/CD with no carrier extension 3 (4) No opinion yet >50 Other Items: ============ Approved minutes from La Jolla Plenary by Aye's. Plans for next meeting. A show of hand points to approximately 50 attendants from this meeting will attend the July Plenary. Will request 1 1/2 days during the week. It is okay if there is an overlap with FDX and T2. Q: Sunday meeting? R: No. Agenda topics: PAR status, Task Force officers, voting rules, Sub Task Force organization, Outline for draft, Technical presentations. Geoff and Howard have an action item to write a meeting report to the SC25 working group. Meeting adjourned 1:15pm.