Date: August 15, 1988 To: X3T9.2 Membership From: Larry Lamers and John Lohmeyer Subject: Minutes of the Electronic Configuration SSWG The meeting convened at 8:00 p.m. on 08/15/88 in the basement of the Antlers Hotel. John listed the alternative proposals on the chalkboard: 1) Probabilistic method using an "Ethernet-like" collision detection method as presented in 88-92 (Nitza) or an RC (one-shot) timeout. 2) A method using additional signals as presented in 88-69R1 (Marazas). Paul Boulay has discussed a similar method on the SCSI BBS. 3) A method using a logically unique function (e.g., embedded serial numbers) as outlined in 88-100 (McGrath). Jerry Marazas stated that SCSI-2 should allow more than one alternative for implementing electronic configuration. He gave a presentation on his revision one document which details a state-machine approach. - optional extension to SCSI - co-existence between old and new devices - signal count reduced to four (eight lines are used for differential) - configuration master can exist at any location on the bus - configuration is allowed at any time - can be extended to more than eight devices on 32-bit bus * each target will poll the four configuration leads and remember the preferred direction depending on which of the two pairs has input * assumption is that only one initiator on bus (errors are handled by operator) * old devices must be pre-configured and their presence is detected by a series of INQUIRY commands prior to the configuration process, new devices do not participate in this. * the configuration master comes up with BSY asserted * a target does not pass through configuration signals until configured. The proposal has not addressed all error conditions yet. Some questions remain to be answered: 1. What is method of global reset if the configuration master powers down, but targets do not? 2. Can re-assignment of addresses be done quickly enough for dynamic tuning? 3. Jerry believes that a trade-off can be made of lengthening configuration time to lower driver current for configuration signals, but the study is not complete. 4. Putting an auto-configuring target on an "old" initiator will not work. It would never respond. Adding a switch seems self-defeating. The SSWG scored the revised proposal according to the set of criteria from the Boise working group meeting (please refer to 88-72): G: Group rating of importance on a 10 point scale with 10 being best. I: IBM original proposal (88-69R0) rating by Boise working group. R: Revised IBM proposal (88-69R1) rating by Colorado Springs SSWG. Changed answers are highlighted with Bold type. Possible Configuration/Termination Assumptions 1. Any device may perform the configuration. G: 10 I: no R: Only one 2. The configuring device may be at any position on the bus. G: 10 I: Yes, with some modifications. R: Yes 3. "New" and "old" devices may be mixed on the bus. G: 9 I: Problem -- old devices do not have a 60-pin cable. R: Yes, but same problem exists 4. The order of new and old devices should not matter. G: 9 I: Problem -- they must be at the end of the cable and that defeats terminator sensing. R: no agreement 5. The solution works for both single-ended and differential drivers. G: 10 I: no -- IBM proposal does not work for differential (Jerry disagrees). R: Yes 6. Add the minimum number of new signals -- This not only saves connector costs, but also saves pins on protocol chips. G: 8 I: Adds 10 pins R: Adds 10 pins 7. New devices should have one or two identical connectors on each device (same sex and wired one-to-one). No "IN" and "OUT" connectors. G: 7 I: no R: no 8. Time to accomplish configuration should not exceed __ seconds. G: 10 seconds I: yes R: yes 9. New devices must remember their assigned address through all but a power outage. They must request a re-configuration when they are powered-on. G: 10 Potential problems -- how to deal with hard reset? I: yes R: Yes for statement one. No for statement two. 10. The solution must be able to accommodate more than 8 devices (for SCSI-3). G: 10 (need up to 32 devices) I: Modified could do 16 devices, 32 devices: no R: yes 11. Powered down devices do not affect the process, including the device that normally does the configuration. G: 10 I: Problem -- could do with a relay. R: Need relay 11b. More than one device can do configuration. G: 10 I: Yes, requires private agreement. 12a. Must detect no terminators, only one terminator, or more than two terminators. G: 4.5 I: Can detect one terminator, assumes initiator is at end of cable. [Jim Schuessler proposed an analog terminator detector which would detect the number of terminators present.] R: Could do analog technique 12b. Terminators are at the end of the cable. G: 4.5 I: No R: No 13. Cabling between devices must be identical. G: 10 I: Yes, with all new devices. R: Yes, with all new devices. 14. In mixed systems of old and new devices, re-configuration must not change any addresses that were previously assigned. Old devices could be very confused by a device that changes addresses. G: 10 I: Yes R: Yes 15. All terminator assemblies must be identical. G: 10 I: Yes R: Yes 16. The solution must be compatible with existing protocol chips. G: 10 I: Yes R: Yes 17. Addressing must be capable of being stable over multiple configurations. (same or different configuring device? Paul Nitza) 17a. via software [to be considered later] R: Yes 17b. via hardware [to be considered later] R: Yes 18. Fifty pins (related to item 6) G: 10 I: No R: No 19. Bi-directional cable G: 9 I: No R: No 20. Must comply with ANSI patent policy. G: 10 I: Yes R: Yes 21. Device can have no SCSI ID to handle more than the maximum number of devices connected to the bus. G: 2 I: Yes R: Yes 22. Possible to daisy-chain with only one connector. G: 10 I: No R: No Review of the revised IBM proposal consumed most of the SSGW's time. The two alternative proposals were reviewed briefly. More time will be devoted to these at the San Jose working group meeting. In Paul Nitza's proposal (88-092R0), any device configures itself if it powers on at a later time. Solution is probabilistic. Problem of old devices powering on late may cause synonyms exists as it does with every proposal. Jim McGrath's proposal (88-100R0), uses a new signal, called "MO". We could use RST with some modifications in its definition. Is re-configuration desirable? Worst case is nine seconds to complete.