X3T9.2/92-113 R1 Date: May 5, 1992 To: X3T9.2 Membership From: John A. Gallant, Chair SCSI-2 Queuing TIB Working Group Subject: April 28 SCSI-2 Queuing TIB Working Group Meeting Minutes John Gallant called the meeting to order at 9:15 a.m. April 28, 1992. The final agenda was as follows: 1. SCSI-2 Queuing TIB [Gallant] (92-075R0) 2. CAM - The Next Generation 3. SCSI-3 Queuing [Penokie] (91-098R7) The following people attended the meeting: Name Organization ------------------------- ------------------------------ Mr. Scott Smyers Apple Mr. Jan Dedek ANCOT Mr. Ed Haske CMD Mr. John Gallant DEC Mr. Ed Gardner DEC Mr. Charles Monia DEC Mr. Ralph Weber DEC Mr. Ted Lappin Exabyte Mr. Robert Liu Fujitsu Mr. Jeff Williams HP Mr. George Penokie IBM Mr. David Lawson Interphase Mt. Jim McGrath Quantum Mr. Jeff Stai Western Digital 14 People Present 1) SCSI-2 Queuing TIB [Gallant] (92-075R0) John Gallant presented the initial goals for creating a TIB around the use of SCSI-2 Tag Queuing. In the discussion around the initial proposal, (92-075R0), there was a very strong push back around theattempt to make ECA and Tag Queuing required to work together. There are a number of vendors, system (PC?) and peripherals that run with Tag Queuing and no formal method for error recovery. The biggest hurdle that this group will have to deal with is "Existing Products". There was a brief discussion around the Auto-sense functionality in CAM and how CA is not sufficient for dealing with error cases when running with Tag Queuing. The problem with using CAM is that there are currently very few implementations of CAM in the market place. Along with this issue the goal of working in a multi-initiator was also criticized. The SCSI-2 document is already broken for multi-initiator and we "shouldn't try to fix it here". The QErr bit, SCSI-2 10h 7.3.3.1, was brought up as being acceptable for handling all possible error conditions, throw all the I/O back up and allow the software on the host deal with the problems. There were two implementation versions that were presented. One was to simply clear the queue and use Request Sense/Unit X3T9.2/92-113 R1 Attention to notify the hosts that all the I/O was gone and the other was to return to the host all the I/O requests with a Check Condition. Of the two implementations the second, each I/O gets a check condition back, was the most acceptable to all the vendors present. The remaining time was spent briefly discussing the possibility that there could be two options, one using QErr and the other ECA, when enabled, to allow further error recovery to occur. There was some limited discussion around using ECA as described in SCSI-2 or having this TIB modify its behavior. The time was limited and the meeting adjourned at 12:00p. We agreed to try and meet again at the working group, in May. It was apparent that there needed to be more system vendors involved in this discussion. For system vendors that do require more interactive error recovery mechanisms and system vendors that run with multi-initiator environments the existing models need to be agreed upon or fixed. There is the explicit dependency on the software "up above" to take care of things. There was not a sufficient number of people in the meeting that could agree as to where the "buck stopped". 2) CAM - The Next Generation There was a brief discussion on features and functionality that the group would like to see in CAM-2. There was no serious discussionsjust the following ideas thrown out: 64 - Bit Data Structures all current 32 bit fields, pointers and counters connection ID scatter/gather maps Regroup fields in the various CCBs Fast/Wide control bits a single transfer parameters bit ? ECA Physical Channel control (Serial) Warm Swap (Description) A scatter/gather map for the targets memory CAM of DADI Expanded functionality for scatter/gather maps a s/g map contains a flags field at the head to define the rest of the map 3) SCSI-3 Queuing [Penokie] (91-098R7) George presented his revision of the day for the SCSI-3 queuing proposal. There was some discussion around what was meant by "media" in the different device types, it was recommended that George look at the SCSI-2 document for examples. It was also recommeded that the two queuing models in the target, single queue vs multiple queue, be made only 1 and single queue is preferred. The ACA functionality was discussed at length, however it's use in error handling and recovery was not. George received a number of comments, both electronicly and during the plenary meetings. He will undoubtably present Rev 8 at the working group meeting.