March 30, 1989 TO: IPI Level 3 Tape Working Group Members FROM: Roger Cummings SUBJECT: CORRECTED TAPE WORKING GROUP MINUTES After the Tape Working Group meeting I distributed the minutes and the proposed changes to the standards to all of the meeting attenders. I also forwarded a copy of these documents for inclusion in the next X3T9.3 mailing. However after reviewing the distributed package several of the meeting attenders pointed out that I had omitted an agreement to define an additional bit in the Attributes Parameter 6E (Slave Reconfiguration) from both the minutes and the changes. Therefore I have amended the minutes accordingly, and a revised version is attached. The revision is limited to the addition of a single paragraph on the penultimate page, and is marked by a revision bar in the left margin. If there are any further corrections required to, or omissions noted from, the minutes I can be reached as follows: Phone: Business (416) 826-8640 x3332 Home (416) 625-4074 (ans machine) Telex/MCI Mail: 650-289-5060 (USA) Fax: (416) 821-6363 Regards ______________________ Roger Cummings Principal Engineer, I/O and Peripherals Systems and Strategies Group Control Data Canada Ltd. 1855 Minnesota Court Mississauga, Ontario L5N 1K7 Canada #yi/rc MINUTES OF THE IPI LEVEL 3 TAPE WORKING GROUP The Level 3 IPI Tape Working Group was hosted by Bob Driscal of Fujitsu America at their facility at 3055 Orchard Drive, San Jose, CA on February 1, 1989. A total of 13 people attended as follows: ASPEN PERIPHERALS John Herron CONTROL DATA Roger Cummings ENDL CONSULTING I. Dal Allan FUJITSU AMERICA Bob Driscal Jim Luttrull IBM Horst Truestedt INTELLISTOR Bob Sheffield LASER MAGNETIC STORAGE Bill Homans Chitu Shah SIEMENS Friedhelm Caprasse STORAGETEK Michael O'Donnell UNISYS Bob Bergey Doug Hennessey Documents were received from both Chuck Stead of Cipher Data, and Gary Stevens of IBM Tucson in lieu of attendance. They were distributed to all participants and are Attachments 1 and 2 respectively. The agenda had been set by Roger Cummings of Control Data Canada, and the first item considered was the presentation on problems with section numbering that had been made by Bob Bergey of Unisys at the October X3T9.3 plenary in Colorado Springs (#88-072). In reviewing the problems it was quickly discovered that they had all been corrected in the ISO version (DIS 9321), and the document editor, Dal Allan of ENDL Consulting, provided a master from which copies were made for all participants. The next item considered was the presentation on problems with the ANSI Level 3 Tape Standard that was made by Roger Cummings at the December X3T9.3 plenary in San Diego (#88-109). Considering the points in their order in the presentation: 1) Attribute Parm 5E It was agreed that this parm is incorrect in the Tape Standards (both ANSI and ISO) and should be corrected to "Both" and "n+1" as an editorial change. 2) Report Addressee Status The command in the ANSI and ISO Tape standards is an earlier version of the command in the Disk standard, and is missing several features. Discussion elucidated the fact that opmod bit 1 had the same function in both versions despite different names (the only difference being that the tape version is more restrictive in requiring the vendor unique parameter to have an ID of 90). Therefore the Tape standard could be updated to the Disk version without breaking any existing implementations - only the Port Query Function would be added. Thus it was agreed that the Tape Standard would be amended to reflect the command in the Disk Standard with the addition of the presently defined Media Status parameter, (and with the Port Mask Parameter corrected to have a length of "n+1".) It was noted in passing that the change to add the Media Status parameter to the Disk RAS command that had been proposed by Bob Bergey and adopted at the Milpitas plenary had already been included in the ISO Disk Standard. 3) Information Transfer Size Override Parm in Read Command The document from Chuck Stead indicated that this parm had not been included due to lack of interest. It was felt that the parameter should be included at the next revision for consistency, and as an editorial change. 4) Master Termination Parameter in Read Command It was agreed that the Disk Standard name is correct and that the Tape Standard should be amended to "Master Termination Permitted Parameter" as an editorial change. 5) Read Error Log Command opmod and extent parameters It was agreed that the situation with this command presents a number of serious problems. The command in the Tape Standard is again an earlier version of the command in the Disk Standard. Although it states that the log contents may be transferred as data, it does not provide the Request Parameter and extent parms found elsewhere to allow the Master to control that process, and is thus incompatible with the scheme described in section 5.2.4 of the Disk Standard. If these parms are included in the command an opmod bit is necessary to define a count in either octets or blocks. This is by convention done by opmod bit 0, which is defined as Clear Error Log in the command version in the existing Tape Standard. Therefore to update the Tape Standard to reflect the current Disk Standard will inevitably break some implementations. After much discussion, it was decided that the problems with the current Tape Standard were serious enough that correction is a must, and therefore it was agreed that the Read Error Log command in the Tape Standard would be replaced by the version in the Disk Standard. It was also agreed that a warning note would be added that stated that a different version had been mistakenly included in a previous revision. With regards to the items identified as discrepancies in #88-109 a) Parm 31 It was agreed that the Disk Standard is incorrect in showing the length of this parameter as "09". It should be "n+1", as the parm would be truncated if a parm 3A is used. This is clearly an editorial change. b) Access Controls command and Access Key Parameter The document from Chuck Stead clarified that this feature had been omitted from the Tape Standard because of lack of interest. The meeting felt, however, that the feature could be useful in the future as tape capacities increase - in particular Dal Allan strongly suspected that it could have application in serpentine devices with track addressability. Therefore it was decided to include the entire Access Permits command from the Disk Standard in the Tape Standard. It was also decided to add a note to the effect that normally in tapes the permits are defined on a partition basis, and therefore the Extent and Data Address parameters are not generally required. Also the Access Key parameter will be added to all relevant transfer commands. Because none of these changes break existing implementations, they can be regarded as editorial. c) Read IPL Command The meeting confirmed that the Read IPL command should be different in the Disk and Tape Standards. d) Attributes Parm 68 Chuck Stead's document again stated that Parm 68 was shown as "Slave" in the Tape Standard because nobody originally involved could conceive of a tape situation in which the Master would have to identify to the Slave the type of peripheral that it has attached. Again though, the feeling of the meeting was that for consistency the same a parameter should be used in both Disk and Tape Standards, along with a note that in the tape situation the Master would not normally send this parameter. d) Attribute Paramater 69 The meeting could see no use for this parameter, and therefore it was decided to remove it from the Tape Standard for compatibility with the Disk Standard. e) Read Command Skip Mask and Boundary Gather Parameters There was some discussion on this subject, mostly about potential uses for the skip mask parameter, but eventually it was decided that no change should be made. This having completed the items contained in #88-109, a call was issued for any other known omissions and corrections. Mike O'Donnell of Storagetek noted the following items: a) Section 6.6 of the ANSI Tape Standard references the Port Control command instead of Path Control. b) In table 16 of the ANSI Tape Standard parameter 50 is shown as "Both" and should be "Master". c) The Port Response and Anticipated Action commands are not referenced in the ANSI Tape Standard. They are however referenced in the ISO Tape Standard. d) For all the commands in Section 7 (excluding subsection 7.2) and Sections 8, 10 and 11 of both the ANSI and ISO Tape Standards, the Major Status code in the Response Packet is shown as "0010" where it should be "0001". e) In the Read Raw Data command described in section 8.2 of the ANSI and ISO Tape standards, opmod bit 1 is not labelled. In the Disk Standards it is labelled as Data Recovery and defined as being a "0". It was agreed after some discussion that it would be best to clearly identify that the bit has no meaning in all situations i.e. mark it as a Don't care. Doug Hennessey of Unisys then raised an issue with the definition of a bit in the Conditional Success parameter, which is defined in the Disk Standard. Octet 2 Bit 1 is labelled "Release of Unreserved Addressee", and this is defined in section 5.4.2.6.15 as "The addressee has received a release command for which there is no reservation". Doug contended that this should clarified to apply to a reservation over the path of this command, as it is not intended that a release command over one path be able to break a reservation made over another. There was agreement on this point from the attenders, and also that the nature of this change was editorial. All problems with, and corrections to, the existing standards having been discussed, the meeting moved on to consider the two enhancement proposals that had been made by Roger Cummings at the San Diego plenary. The first proposal discussed was the Threshold Read command (#88-110). Roger described the background to the proposal in detail, and several of the attenders expressed surprise that IPI-3 Tape included no analogue to the FIPS SILI (Suppress Incorrect Length Indication) mode. The proposed threshold scheme was agreed to be superior due to its ability to break chains, but there was also support for the SILI scheme. Therefore it was agreed that support for both modes should be included. Roger explained that the particular implementation proposed had been chosen to closely mimic Control Data's existing scheme that used a Vendor Unique parameter. There was general feeling, however, that the proposed implementation was somewhat obtuse, and could be substantially improved. There was also sentiment that both the Threshold and SILI modes be made available on all transfer commands, and not just on Reads as Roger proposed. Thus it seemed that an implementation using a common parameter would be optimal, and the Transfer parameter (3C) was agreed as being the best. Further details were postponed for later discussion. The last subject considered was the Deferred Error proposal (#88-108). Roger again gave a background to the proposal, which was the desire to allow cartridge tapes to be used in journaling applications that are presently restricted to the use of unbufferred reel to reel tapes. A number of the attenders expressed the view that the proposal was covered by the existing definition, but a detailed review of section 4.5 gave rise to some doubts - mostly because the sentence that described the use of an Asynchronous Response included the verb "shall". Dal Allan was of the opinion that one of the major advantages of IPI had been negated. Roger emphasized that he was not trying to invalidate the existing implementation but merely trying to add an alternative that had merit in some situations. | It was suggested that it was quite feasible that a slave be | designed to support both methods of handling Deferred Errors, and | that in that event it would be desirable to allow the Master to | select which method is used. A reserved bit in Attributes | Parameter 6E (Slave Reconfiguration) was agreed as being the most | logical place for this feature. There was then an extended discussion regarding the details of an implementation that would support the Deferred Error proposal. The discussion centered on the depth of command queue required, and alternate queue structures. Bill Homans of LMSI in particular considered possible implementations in depth before agreeing that the approach was feasible. Friedhelm Caprasse of Siemens noted the use of the Inhibit Operation Response on Success feature in the proposal, and suggested that it would be useful in a number of situations to be able to control the feature on a per command basis. At present the feature is controlled by parm 50 of the Operating Mode command. After discussion it was agreed that the per command approach would be generally useful. Therefore the decision was made to add add all of the features found in Op Mode parm 50 to the Transfer parameter (3C), and to add 3C to the parameter list for all read-type transfer commands (all write-type commands have 3C today). Thus the existing Transfer parameter of length "02" is replaced by one of length "04" with the following layout: Octet 1 Bits 7-0 as current definition Octet 2 Bit 7 Threshold Mode Bit 6 Suppress Incorrect Length Indication Bits 5-0 reserved Octet 3 Bit 7 Post Conditional Success if Error Retry Bit 6 Post Conditional Success on Data Correction Bit 5 Inhibit Operation Response on Success Bit 4 Inhibit Extended Substatus Response Bits 3-0 reserved This concluded the business of the meeting. Bob Driscal was thanked by all attenders for providing the excellent facilities for the meeting.