Date: September 6, 1988 To: X3T9.2 Membership From: Larry Lamers, SCSI-2 General Editor Subject: August 1988 SCSI-2 Working Group Meeting Minutes Jim McGrath of Quantum hosted a working group meeting August 29-31 in San Jose, CA at the Radisson Hotel. I would like to thank Jim and Quantum for hosting this productive meeting. The final agenda was: Agenda Items for SCSI-2 1. [] bracket removal 2. Review of S2R5 3. LOG SELECT/SENSE review 4. ECA via AEN Issue 5. RECONNECT QUEUE Message 6. Processor Device Model [88-064] (Stai) 7. AEN Revision [88-065R2] (Boulay) 8. Rotational Position Locking 9. Medium Changer Devices 10. SCSI Device Connector Orientation [88-80R1] (Marazas) 11. Multiple IDENTIFY messages [see 9.1] (Snively) 12. Bull's comments on single-ended [88-98] (Allan) 13. 12-byte CDB's for LOG SENSE/SELECT [88-95] (Pentecost) 14. Open pins (Bob & Ed need to come to consensus) [88-77] 15. DEC open pin and terminator power issue [88-48R1, 88-77, 88-87] Agenda Items for SCSI-3 50. SEND DIAGNOSTIC Pages [87-186] (Spence) 51. LOAD SKIP MASK command proposal [87-203] [87-217] (Floryance, McIntyre) 52. More that 8 devices on wide SCSI [87-206] (Harms) 53. Search Command modifications [88-002] (Stai) 54. Expanded RelAdr Bit Definition [88-007] (Boulay) 55. Autoconfiguration SSWG [88-69R1, 88-92, 88-100] (Marazas, Nitza, Boulay) SSWG Schedule: Monday 8:00 a.m. - 9:00 a.m. Hierarchical Filemarks (*) Monday 7:00 p.m. - ? Auto-configuration Tuesday 9:00 a.m. - noon Transmission Line issues Tuesday 7:00 p.m. - ? Auto-configuration Wednesday 8:00 a.m. - 9:00 a.m. Medium Changers Wednesday 1:00 p.m. - ? Medium Changers (*) A study group was formed on hierarchical filemark structures for sequential-access devices. Bill Homans is leading the group. Please contact him at 215-666-4764 for further information. The following people attended the meeting: Name Status Organization ------------------------------ ------ ------------------------------ Mr. Robert N. Snively P Adaptec, Inc. Mr. Charles Brill P AMP, Inc. Mr. Bob Whiteman A AMP, Inc. Mr. Chuck Fannin A Archive Corp. Mr. Ken Post A Archive Corp. Mr. James J. Semenak P AT&T Mr. Christopher J. Rhea O AT&T Mr. Gregory Fry A Ballard Synergy Corp. Mr. Dan Davies P Cipher Data Products, Inc. Mr. Robert Kellert P Cirrus Logic Inc. Mr. Gene Milligan P Control Data Corp. Mr. Gerald Houlder A Control Data Corp. Mr. Bart Raudebaugh P Cygnet Systems Mr. John Morse A Digital Equipment Corp. Mr. Gerald Maurer P Emulex Corp. Mr. Rick Muething V Emulex/Asix Systems Ms. Jean Kodama V Emulex/Computer Array Dev.Inc Mr. I. Dal Allan P ENDL Mr. Robert Liu P Fujitsu America, Inc. Mr. Bob Pentecost O Hewlett Packard Co. Mr. Kurt Chan O Hewlett Packard Co. Mr. Andy Hospodor O I/O XEL Mr. George Penokie P IBM Corp. Mr. Gerald Marazas A IBM Corp. Mr. David A. Buesing O IBM Corp. Mr. Erik Walberg O Konica Technology Corp. Mr. Ken Bartlett V Konica Technology Corp. Mr. William Homans P LMSI-CPI Mr. Robert Johnston S LMSI-CPI Mr. Paul Boulay A LMSI-OSI Mr. Skip Jones A Maxtor Corp. Ms. Donna Pope O Maxtor Corp. Mr. Dale Smith V Maxtor Corp. Mr. Kim Ryal A Miniscribe Corp. Mr. James Schuessler P National Semiconductor Mr. Thai Nguyen V National Semiconductor Mr. Jerry Fredin V NCR Corp. Mr. Dennis P. Trupski P Olivetti Mr. Charles Yang P Panasonic Industrial Co. Mr. James McGrath P Quantum Corp. Mr. Dexter Anderson P Scientific Micro Systems Mr. Lawrence J. Lamers P Sony Corp. of America Mr. David A. Hatch P Stewart Connector Systems Mr. Harvey Waltersdorf P Thomas & Betts Mr. Jeff Stai P Western Digital Status Key: P - Principal A - Alternate O - Observer S - Special Interest (Does not get mailings) V - Visitor The following new documents were distributed at the meeting: Document Doc Date Author Description of Document ------------- -------- --------------- --------------------------------------- X3T9.2/88-65 8/29/88 P. Boulay Asynchronous Event Notification Rev 2 Revision X3T9.2/88-69 8/29/88 G. Marazas Physical Architecture -- SCSI-2 Rev 2 Interface X3T9.2/88-80 8/29/88 G. Marazas SCSI Device Connector Layout Rev 2 X3T9.2/88-103 8/26/88 D. Davies Generic Error Recovery Mode Page X3T9.2/88-104 8/29/88 A. Hospodor Reporting Disk Rotation Speed X3T9.2/88-105 8/29/88 G. Fry Direct-Access Device Model X3T9.2/88-106 5/24/88 J. Schuessler Electrical Test Results X3T9.2/88-107 B. Homans Save Set Mark (a.k.a. Hierarchical Filemark) Proposal X3T9.2/88-108 8/31/88 R. Snively B Cable Data Transfer Requirements 1) Review of square brackets: See general editors master S2R5 document for details. 2) Review of S2R5 -- Paul Boulay moved, and Gene Milligan seconded that I_O should be changed to I/O to more correctly indicate that it is an input/output process. Motion passed. The feeling is that with underscores for new text I_O was not clear. -- The working group unanimously agreed to delete references to shield ground in section four. -- The CHANGE DEFINITION command engendered a fair amount of discussion. Jeff Stai wanted to limit the implementation so that one definition (the currently specified one) applied to all initiators. Jeff planned to worked up a proposal on this subject. -- It appears that origin of Section 8 in S2R5 is suspect since it looks like rev 5 came from rev 3 not rev 4. 3) LOG SELECT/LOG SENSE review This discussion was fairly benign except for 12-byte CDBs. See item 13. 4) ECA via AEN Issue - See Item 7. 5) RECONNECT QUEUE message The UNORDERED QUEUE TAG message will be renamed to QUEUE TAG message and used in a bi-directional manner. The ORDERED QUEUE TAG and HEAD OF QUEUE TAG messages will be outbound only. There was a request to to organize the message table like the text is, i.e., all the queue messages grouped together. 6) Processor Device Model Jeff Stai has not completed this so the item is carried over. 7) AEN Revision Paul Boulay's proposal was reviewed and commented on. He will prepare yet another revision that cleans up all the concerns and conflicts of AEN. 8) Rotational Position Locking Larry Lamers and Bob Mortensen have not completed this so the item is carried over. 9) Medium Changer Devices The SSWG on this subject met several times. They are working on another revision of section 16. Paul Boulay is editing this section; contact him for the latest information. 10) SCSI Device Connector Orientation Gerald Marazas presented a revised connector orientation diagram. See 88- 80R2. Gene Milligan did not like the wording on the power zone. A long discussion ensued pitting the buyers against the builders. Neuterization and genericization was performed so that a consensus be reached. A straw vote was taken to recommend a revised version of 88-80 will be incorporated in an appendix. The vote was 10 in favor, 10 opposed. A straw vote was taken to have a small illustration in figure 4 to show connector orientation. The vote was 8 in favor, 1 opposed. 11) Multiple IDENTIFY messages The sentiment was not to explicitly prohibit multiple IDENTIFY messages, but to put in an implementors note explaining the need for architectural purity in an effort to lower their use. 12) Bull's comments on single-ended - See Item 15. 13) 12-Byte CDBs for LOG SELECT/SENSE This was a fooler as often happens in this committee. It should have been a simple request as others of this nature have been. Jim McGrath wanted a block method of doing this, Jim Semenak was opposed to 12-byte CDBs, and Bob Pentecost did not express a reason for wanting it. Dal requested that it be taken off-line. Jim Semenak will lead a SSWG to develop a proposal for 12- byte log commands that will deal with the nitty-gritty issues. The 10-byte version of the commands will be deleted and the page headers revised to deal with people who have megabytes of log data to return. 14) Open Pins John Morse wants to defer consideration of adding more terminator power lines until next plenary so that proper documentation can be made available. Dal requested that the following items be addressed: current requirements, parallel current paths what happens when one path is interrupted, UL, CSA requirements and justify the need for more terminator power lines. 15) DEC open pin and terminator power issue This item was the jumping off point for the transmission line SSWG. What follows is as much of the meat of the conversation as I could record. This is a important subject matter and I may not have gotten all the points or may have not recorded the information correctly. Please review it, add to, and correct it, as I would like these minutes to bring the technical issues to light. Jeff Stai conducted the Transmission Line SSWG. The charter was to explore everything, including 25 mil cable. Jim Schuessler wanted to set objectives, stick to the facts, let companies decide on the risk they want to take. Dal Allan suggested the group make recommendations on terminator power, cable gauge, and termination. Gene Milligan presented overheads on testing results of six-meter single- ended with protocol chips. Control Data uses six different manufacturers of protocol chips. He had test data on synchronous SCSI, in particular the phenomena of REQ/ACK at the transmitter end. The data suggests that more than termination and reflections are worries; the chips and the distortions they introduce are also of concern. Not tested to worst case. [need to get overheads from Gene.] The first configuration tested was: Control Data's TWD host adapter, 6" flat, connector, 3 m round shielded cable, connector, .5 m twisted pair, drive, .5 m twisted pair, connector, 2 m twisted pair, drive. The second configuration tested was: Terminator, 6" flat, host adapter, 6" flat, connector, .5 m round shielded, connector, 2 m round shielded cable, connector, 6" flat ribbon, .5 m twisted pair, drive, .5 m twisted pair, connector, 2 m twisted pair, drive. One configuration was run at 3.9 mbs, one at 4.0 mbs, for 16 hours of continuous data transmission. No errors were reported. The overheads show discontinuities in the REQ/ACK signal, seen as two REQs by host. The tests were done using a four-volt term power source. The overheads showed reflection and spikes from protocol chip switching. Gene concluded that this demonstrates that using poor cabling practices, and six-meters of cable you can operate error free for 16 hours. Jim Schuessler said test set up was only three devices, and they were far apart; the real worst case occurs if you have more devices and they are grouped closely together at a distance from the host (i.e., the targets are clustered). Bill Homans said if the terminator power voltage is kept up, the glitch is not a problem, but at four volts it is. The Emulex transmission-line expert stated that what is needed is equal overshoot and undershoot and that the Thevinin equivalent must match cable impedance. One must balance the driving impedance, and then select a voltage to equate overshoot and undershoot, and drive it with the proper current. Dal stated that before we pursue options we need to decide what are degrees of freedom we have, in other words, what are we not willing to change in order to solve the problem. The sentiment is that changing Vterm is the only thing likely to be accepted by the committee. Dal concluded that proposals already submitted deal with the one option that has some promise. Bob Snively thought Control Data's glitch would disappear if the cables were really short. We must consider the terminator power signal as a transmitter. The glitch then is really a transmission line problem not a chip problem. The glitch is worse on fast chips. Jim Schuessler presented his test results. Reflections are biggest problem due to mismatch of cable impedance and terminator impedance. In SPICE model simulations with clustered devices, a significant impedance discontinuity occurred at clustered devices. At least one foot separation is needed to have any improvement. Shorter cables help, evenly spacing devices on the bus also helps. Using garden variety protocol chips would increase cross-talk, but not affect the reflection problem. Dal reported that systems integrators are finding that beyond five devices the error rate increases, especially with longer cables to clustered devices. Low impedance cables help the clustering problem, but requires higher driving current. ECL parts could help this. The document should contain a requirement that 90 to 132 ohm cable be used for connecting devices. The what ifs: -- what if we went to 100 ohm termination? -- need higher current drivers, use existing 100 ohm cable. -- older drivers would not work with new terminators. -- new drivers would work with old terminators. -- what if we change cable impedance from 90-132 ohm? -- there is so little noise margin that changing the cable isn't feasible. -- Dal has action item to develop the electrical characteristics for round shielded cable and include them in SCSI-2 R6. -- what if we change the Vterm power spec? -- increasing voltage increases reflections, and increases sink current. -- overshoot and undershoot increases, problem is on negation not assertion. As the negation problem gets better, the assertion problem gets worse. Negative undershoots are often a board layout problem, undershoot is typically a result of inductance. Jeff Stai asked that a (any) semiconductor person explain how present CMOS designs deal with undershoot. Bob Kellert suggested that we possibly just change REQ/ACK to high current transceivers / termination, since these signals are high speed and edges are important. He wants to keep REQ/ACK and SEL from going open drain ever, so he can use totem pole drivers. The 28AWG wire has 50-70 ohm typical impedance, and 25 mil centers require a max of 28AWG, so 28AWG on 25 mil centers does not meet acceptable cable specs. The resulting problem is that the pinouts in S2R5 are based on a 25 mil center flat cable. Therefore the pinouts may need to be changed. A pitch-switcher would increase stub length by about two inches. Increasing stub length is not recommended by Jim Schuessler since it increases the pulse distortion. So using a pitch-switcher would decrease the available stub length in the cabinet. Using a 50 mil cable back-end on a high-density connector is feasible, however the signal pairing going through the connector into the pc-board would be lost, since the cable is folded. Dave Hatch presented information on cabling to connectors that showed the cable being terminated to the connector while maintaining signal pairing. It uses a dual level termination with 50 mil cable. Bob Whiteman has action item to look at the 50-mil 28AWG cable daisy-chain needs by designing a new back-end for the high-density connector. Transmission line SSWG recommendations: 1) All cable used in the system shall be 90-132 ohm. 2) Reduce the 6 m length. 3) Provide advice on cable lengths in implementations. Gerry Fredin accepted an action item to create wording for this. 16) Rotational Speed field in mode parameters [88-104R0] (Andy Hospodor) This is a request for a way to report rotation speed in MODE SENSE. The reasons for this request are for testing and to better optimize sector skews. Jeff Stai had some comments that the SCSI controller should know what the best skew is. Bob Snively had an almost traumatic reaction to Andy's comments on poor testability of SCSI. Straw poll of adding rpm; 11 in favor, 5 opposed. Table 8-57 needs a continue statement. 17) Direct Access Device Model. [88-105] (Greg Fry) Dal stated that Bill Burr has completed a model for direct-access devices. However no one except Gene Milligan has any recollection of every seeing it. Dal requested that John Lohmeyer provide this model in the next mailing. [John found this note too late to obtain Bill's model for the mailing. jbl] Greg will put his model up on the SCSI BBS for comment. 18) Clarification of B Cable data transfer requirements [88-108] (Bob Snively) There was an involved discussion, mostly centered around how forgiving hosts and controllers should be. Paragraph 3 will be modified so that if an error does not occur, the device will wait until the REQs & ACKs agree again. Paragraphs 1, 2, and 4 will be modified per the ensuing discussion. Bob will revise the proposal. 19) Generic Read-Write Error Recovery Mode Parameters Page [88-103R0] (Dan Davies) Straw poll on moving the Read-Write Error Recovery page from section 8 to section 7. 12 in favor, 5 opposed. There was strong opposition from some members of the group who felt that initiators must deal with mode parameters on a per device type basis and do not want limitations of a disk implementation. Dan has an action item to neuter the page and make it totally generic. 20) Hierarchical filemarks for tape Bill Homans read a report on hierarchical filemarks. The SSWG is in agreement on the additional wording needed (see 88-107R0). A further refined proposal will be developed prior to the plenary meeting. 55) Autoconfiguration SSWG - Paul Boulay gave the report from the SSWG held on Monday evening. There was a very low level of interest and the group met over dinner. Jerry Marazas had not yet arrived. Gerald Marazas brought a revised proposal on autoconfiguration. It was discussed at the Tuesday evening SSWG. Level of interest in the Tuesday evening autoconfiguration SSWG was also low. There were three persons interested besides Paul Boulay who lead the SSWG. Jim McGrath and Gerald Marazas each have an action item to prepare a document that details the issues in the SCSI-2 document that have to be dealt with in order for their respective autoconfiguration proposals to be smoothly incorporated in SCSI-3.