@Date: July 12, 1989 To: X3T9.2 Membership From: Lawrence J. Lamers, X3T9.2 Recording Secretary John B. Lohmeyer, X3T9.2 Chairman Subject: July 10-11, 1989 X3T9.2 Working Group Meeting Joe Lawlor of AT&T hosted this meeting in Chicago, Illinois at the O'Hare Ramada Inn. The working group wishes to expresses its sincere appreciation to Joe for hosting this meeting in a room with an interesting, albeit distracting, view. Since X3T9 forwarded SCSI-2 to X3 for public review, this meeting focused on SCSI-3 issues that had previously been deferred. The final agenda was as follows: 1. Additional SEND DIAGNOSTIC Pages (87-186R0) [Spence] 2. LOAD SKIP MASK Command Proposal (87-203 & 87-217) [Floryance, McIntyre] 3. More Than 8 Devices On Wide SCSI (87-207) [Harms] 4. Search Command Modifications (88-002) [Stai] 5. Expanded RelAdr Bit Definition (88-007) [Boulay] 6. Autoconfiguration SSWG (88-69R1, 88-092R0, & 88-100R0) [Marazas, Nitza, McGrath, Boulay] 7. Error Handling Action Codes (88-127) [Penokie] 8. Autosense [Nitza] 9. Request For Data DAT Device Type [Andrews] 10. Documentation Layering [Stephens] 11. Single-Cable 16-bit Wide SCSI [Penokie, Lohmeyer, Lamers, Stai] 12. Alternate Physical Layers (e.g., fiber optics) 13. LOGICAL UNIT RESET Message Proposal (89-058) [Lohmeyer] 14. Arbitration Fairness Proposals (89-61 & 89-65) [Penokie, Buesing] 15. Log counter definitions (89-96) [Houlder] 16. Editing the SCSI-3 document 17. CD-ROM READ SUB-CHANNEL (89-95 & 89-98) [Boulay, Lin] 18. Comments on SCSI-2 Revision 10 19. SHADOW WRITE and SHADOW READ commands The following people attended the meeting: Name Stat Organization ---------------------- ---- -------------------------- Mr. Robert C. Herron A 3M Company Mr. Al Wilhelm P Adaptec, Inc. Mr. Thomas Newman V Adaptec, Inc. Mr. Scott Smyers A Advanced Micro Devices Mr. Joe Lawlor P AT&T Mr. James J. Semenak A AT&T Mr. Bill Duran V California Peripherals Mr. Bharat Shah A Cipher Data Products, Inc. Mr. Robert Kellert P Cirrus Logic Inc. Mr. Joe Chen A Cirrus Logic Inc. Mr. Paul Hanmann P Emulex Corp. Mr. I. Dal Allan P ENDL Mr. Robert Liu P Fujitsu America, Inc. Mr. Kurt Chan P Hewlett Packard Co. Mr. George Penokie P IBM Corp. Mr. Gary R. Stephens A IBM Corp. Mr. David A. Buesing O IBM Corp. Mr. Gerald Houlder A IMPRIMIS/CDC Mr. W. Dean Wilson P JAE Mr. Paul Boulay A LMS-OSD Mr. Lawrence J. Lamers P Maxtor Corp. Mr. Stephen Cornaby P Micropolis Corp. Mr. John Lohmeyer P NCR Corp. Mr. Bill Medlinski A Panasonic, MECA-BEC Ms. Rita Lin P Sony Corp. of America Mr. Mike Yokoyama V Sony Corp. of America Mr. Robert Monsour V Stac Electronics Mr. Robert N. Snively P Sun Microsystems, Inc. Mr. D. W. Spence P Texas Instruments Mr. Jeff Stai P Western Digital Mr. Doug Pickford A Western Digital ====================== ==== ========================== Count: 31 Status Key: P - Principal Member A - Alternate Member O - Observer V - Visitor The following new documents were distributed at the meeting: Document Doc Date Author Description of Document ------------- -------- --------------- --------------------------------------- X3T9.2/89-94 5/5/89 G. Penokie 16 Bit Data Path on a Single 68-pin Connector X3T9.2/89-95 7/7/89 P. Boulay Read Subchannel Command Revision X3T9.2/89-96 7/7/89 G. Houlder LOG Counter Definitions X3T9.2/89-97 7/8/89 G. Houlder 16-bit SCSI-3 Protocol X3T9.2/89-98 7/6/89 R. Lin Sony "No" vote on SCSI-2 Forwarding Motion Results of meeting 1. Additional SEND DIAGNOSTIC Pages (87-186R0) [Spence] Bill Spence requested that this item be deferred until X3B7 completes its proposal. Dal Allan commented that X3B7 lacks representation from X3T9.2 in developing their draft standard. Also X3B7 is basing its proposal on SCSI- 1, not SCSI-2. 2. LOAD SKIP MASK Command Proposal (87-203 & 87-217) [Floryance, McIntyre] There are two schools of thought on how this function should be accomplished. There are some vendor unique implementations already. IPI-3 defines the LOAD SKIP MASK approach while others think a scatter write and gather read approach is better. Bob Snively commented that with the low overhead of modern targets this kind of command may no longer be needed. Dal Allan countered that existing operating systems often have more overhead than the target and may still need this function. George Penokie suggested that the committee should not pursue the issue at this time. Dave Buesing, also of IBM, thought the proposal should be pursued. Dal Allan said that flat-space operating systems, those that address RAM and disk space in a contiguous address space, use scatter/gather operations to improve performance. George Penokie stated that IBM would withdraw their current proposal (87- 203) and, if necessary, a new or revised proposal will be submitted for consideration. 3. More Than 8 Devices On Wide SCSI (87-207) [Harms] While this proposal was originally submitted for the A and B cable approach to wide SCSI (hereinafter referred to as the twin cable approach), Dal Allan suggested that it should be tied to the single-cable 16-bit proposal (see agenda item 11). Dal commented that the market has not enthusiastically responded to the twin cable approach in SCSI-2. However, the 16-bit proposal has had a high degree of interest. Also further complications that might ensue from adding more than 8 devices to twin cables could confuse implementors. John Lohmeyer commented that in simulations done at NCR the bus could support even more than 32 SCSI-2 devices in some applications. He recommended that the committee not impose any arbitrary limits on the design. Jeff Stai felt that implementing 16 devices in a 16-bit chip could result in some differences and compatible issues if both the twin-cable and the single-cable 16-bit approaches were adopted. Dal also commented that the fairness issue becomes more important if more than 8 devices are permitted (see agenda item 14). Action was deferred on this issue pending more work on the single-cable 16- bit proposal and in hopes that David Harms of AT&T would be able to attend a later portion of the meeting to more clearly define the AT&T requirements. 4. Search Command Modifications (88-002) [Stai] This proposal actually originated from Paul Nitza last year when the LOAD SKIP MASK command was proposed. Jeff Stai recommended that this proposal be dropped until a LOAD SKIP MASK proposal is resurrected. Dal Allan said that while he is not particularly interested in SEARCH commands, the committee should consider adding the IPI-3 SEARCH command feature that allows for powerful boolean constructs. 5. Expanded RelAdr Bit Definition (88-007) [Boulay] While the hardware limits of 1982 have subsided and expanding the RelAdr bit definition has utility for optical drives, Paul Boulay volunteered to withdraw the document. 6. Autoconfiguration SSWG (88-69R1, 88-092R0, & 88-100R0) [Marazas, Nitza, McGrath, Boulay] Bob Snively requested that we abandon the current proposals, but keep the subject because the concept has merit. Others thought it is more appropriate to review the existing documents and developing a new proposal. The two deterministic approaches previously identified work either by: 1) using additional signals which are not bussed and thus establish unique device identification through their physical position on the cable 2) using an unique device identification which would require a manufacturer code, product code, and serial number. A non-deterministic approach has also been identified that uses a pseudo- random sequence to identify devices -- this approach has some non-zero probability of failing to find unique SCSI device IDs. Paul Boulay graciously volunteered to return control of this Special Subject Working Group (SSWG) to the general working group. 7. Error Handling Action Codes (88-127) [Penokie] George Penokie declined to defend the current proposal any further. He said that he would re-submit this proposal if IBM still requires these codes. 8. Autosense [Nitza] While the previous autosense proposal transferred the sense data in an extra DATA IN phase, there was considerable interest in doing this function by extending the STATUS phase. Bob Snively accepted an action item to develop a proposal for extended status within the next three months. The tentative thoughts were: 1) that the extended STATUS phase would still be transferred asynchronously 2) it would be enabled by a message and/or a command 3) the length would be adjustable at the initiator's option through mode parameters 4) the acceptance of extended status would clear the contingent allegiance. Some issues remain: 1) Should the enabling method be a message or command? 2) Should the format of the extended status be identical to the REQUEST SENSE data format? or shorter? 3) Should implementation be mandatory or optional? 9. Request For Data DAT Device Type [Andrews] Deferred pending a proposal from Harlan Andrews. 10. Documentation Layering [Stephens] Deferred pending a submittal (89-036) from Gary Stephens. 11. Single-Cable 16-bit Wide SCSI (89-94) [Penokie, Lohmeyer, Lamers, Stai] Considerable time was spent on this agenda item. George Penokie's document identifies many of the changes needed to achieve a single-cable 16-bit SCSI. In many cases, the document has fill-in-the-blanks. A few errors were found during the discussion, so George agreed to correct the errors and provide a revision 1 to the proposal. George's requirements and wants were summarized as follows: 16-bit wide data path \ 1 high-density connector > Penokie Requirements 1 cable, not 2 / 10 Mtransfers/sec \ 25 meters > Penokie "want-a-sees" fully integrated / superior electrical characteristics \ (comparable to differential) > Other possible benefits practical cable (70 - 90 ohm) / George agreed to reverse his arbitration priority so that 8-bit devices would not fail in a 16-bit environment. He also pointed out a few other errors that he intended to correct in revision 1 of 89-94. The chip vendors were asked whether a 16-bit single-ended protocol chip could be fitted into one chip. They agreed that the answer is yes. Today's differential drivers cannot be highly integrated, however. Several people indicated that the industry is beginning to request differential drivers and receivers for more applications. Unfortunately, differential is often impractical for small form-factor devices. There is interest in finding a driver/receiver solution that gives the electrical benefits of differential, but can be integrated in the protocol chip. Bob Snively suggested using differential receivers with single-ended drivers. It is actually an improved single-ended receiver, with a settable, lower switching point and narrower range. Bob suggested that this may eliminate ground shift issues and bring the threshold voltage down to 1.2 v. There was some skepticism over whether this approach would really work much better. The issue was taken off-line. Another possibility mentioned was a reduced-power differential. Several interesting thoughts were expressed, but they all need further development. There was also a discussion of command set issues that would result from increasing the number of devices permitted on the bus. There would be impacts on the copy commands and the reservation commands. Some three-bit fields exist for the SCSI device number. These would have to be increased to at least 4 bits and it may be desirable to increase them to 5 bits, anticipating a need for 32 devices on a 32-bit cable. John Lohmeyer also suggested increasing the logical unit fields in these commands anticipating a need to increase the number of logical units permitted per SCSI device. Dal Allan suggested that perhaps we should simply abandon the copy commands. They could continue to be supported, but only by devices with device addresses less than 8. 12. Alternate Physical Layers (e.g., fiber optics) Members of X3T9.2 were encouraged to attend the Fiber Channel meetings to support the SCSI position. 13. LOGICAL UNIT RESET Message Proposal (89-058) [Lohmeyer] Deferred until the September working group meeting. 14. Arbitration Fairness Proposals (89-61 & 89-65) [Penokie, Buesing] George Penokie gave a thumbnail sketch of his proposal. Basically, SCSI devices note which other devices participate in each arbitration phase. Higher priority devices refrain from arbitrating again until the lower priority devices have had a chance to win arbitration. It is a hardware based scheme, compatible with existing devices. It does not violate the current standard. Current designs would need revision to implement this scheme effectively. Dave Buesing gave a thumbnail sketch of his proposal. It is both a hardware and software based scheme. Only one initiator is permitted and it must have SCSI device address 7. When a target arbitrates for the SCSI bus, the initiator puts its ID on the bus near the end of the arbitration delay. Thus the initiator always wins arbitration. During arbitration, the initiator captures the IDs of the targets that participated in the arbitration phase. It then selects the target to which it prefers to communicate. Dave uses the NOOP message at this point to inform the target that it should proceed as if it had won arbitration and had reselected the initiator. This effectively converts SCSI into a master-slave interface and gives the master control over the order that it communicates with targets that are ready to transfer data. It was pointed out that Dave's proposal is incorrectly labeled a fairness proposal -- it is actually an unfairness proposal. The initiator is given complete control of the bus. It might be stretched into working in a multiple initiator environment, but not cleanly. The two proposals are not competitors. Each solves a different problem. If a fairness protocol becomes part of the document it is intended to go in section 5. 15. Log counter definitions (89-96) [Houlder] Gerry Houlder pointed out that the SCSI-2 document does not define the error log counters beyond giving them a name. He provided a "quiz" to help define the counters. The working group agreed that the counters were not defined, but it was reluctant to define them. Efforts to precisely define what is an error have often met with stalemate. Instead, the working group agreed that an implementors note should be added to section 7 say something like: "IMPLEMENTORS NOTE: The exact definition of the error counters is not part of this standard. These counters should not be used to compare products because the products may define errors differently." 16. Editing the SCSI-3 document John Lohmeyer pointed out that editing the SCSI document has become an enormous task. Also, publishing and distributing a 600 page document is expensive. Finding competent editors with support from their management to devote a significant portions of their time to editing the document is getting more difficult. He was looking for suggestions on ways to overcome these problems in SCSI-3. While the SCSI-3 document could be layered or have sections published separately, several people strongly felt that the standard be published as one document and not be separate standards. The idea of hiring a full-time or part-time editor was explored. How to collect the funds to pay the editor and how to judge the quality/acceptability of the editor's work may be difficult. John Lohmeyer accepted an action item to discuss these issues with Del Shoemaker. The ISO conversion task of SCSI-2 was considered as a separate task from the SCSI-3 general editorship. 17. CD-ROM READ SUB-CHANNEL (89-95, 89-98) [Boulay, Lin] Paul Boulay present three possible solutions to the Sony problem. He favored doing either option 1 or option 3 (possibly later). Paul was opposed to option 2 (the Sony proposal), because he felt it substantially alters the meaning of the command. Rita Lin felt that Paul's option 1 is insufficient. Part of option 1 needs to be done in any case, but she preferred option 2. She was willing to compromise on doing option 3 plus a portion of option 1. The working group requested that Paul and Rita continue to work on a mutually acceptable solution to the problem. 18. Comments on SCSI-2 Revision 10 The LUNTAR and LUNTRN definitions on page 11-6 need to be added to indicate that they are the source logical unit not the destination logical unit as defined in the IDENTIFY message. Larry Lamers plans to convert the SCSI-2 document to ISO style and perhaps to another word processor during the public review period. He will make the above fix as well as several editorial changes that were pointed out during the meeting. 19. SHADOW WRITE and SHADOW READ commands John Lohmeyer indicated that some people at NCR are interested in defining a command like the IPI-3 SHADOW WRITE command. A SHADOW READ command is not strictly necessary, but could improve performance. Dal Allan suggested that the IPI-3 commands be adopted. John Lohmeyer said he may bring in a proposal on the subject and invited anyone else interested in this function to submit a proposal.