FC-Tape December 98 Minutes

Stewart Wyatt stewart at hpbs3928.boi.hp.com
Fri Dec 18 13:40:32 PST 1998

* From the T10 Reflector (t10 at symbios.com), posted by:
* Stewart Wyatt <stewart at hpbs3928.boi.hp.com>

     Fibre Channel Tape Profile Meeting                 T11/98-619v0
     Tucson, Arizona, 16 December 1998
     Stewart Wyatt, Secretary
     1. Introductions: Chairman Dale Lafollette (STK) called the meeting 
     called to order at 12:35 PM and had the participants introduce 
     2. Approval of this Agenda: T11/98-592v0
     The agenda was approved with the addition of a new item, 6.1 FCP-2: 
     Bob Snively
     3. Approval of 11/3 Minutes: T11/98-567v0
     Minutes were approved.
     4. Review of Old Action Items:
     Secretary Stewart Wyatt, HP, reviewed the action items.
     #1 Stewart Wyatt will get a resolution on HP SCSI command comments. 
     Resolved in HP letter ballot comments.
     #2 Erich Oetting will propose to the T10 working group that the Write 
     Exclusive Persistent Reservations be made illegal for SSC devices. 
     Erich was not present, deferred to next T10 meeting.
     #3 Erich Oetting will propose to the T10 working group that the 
     communications devices command set be made obsolete. Erich was not 
     present, deferred to next T10 meeting.
     #4 Dave Peterson will attempt to document EOD behavior more precisely 
     in SSC. Dave has posted a new version of the SSC, deferred to the next 
     T10 meeting.
     #5 Dave Peterson will enter a letter ballot to correct the I5 IU 
     specified in FC_Tape. Completed.
     #6 Bob Snively will take a recommendation to the T10 working group to 
     obsolete the Flag bit in the Control byte. Completed.
     #7 Bob Snively will enter a letter ballot comment to make FC-Tape 
     compatible with the changes in FCP-2 regarding handling of CRN's. 
     ( and Annex D.) Completed.
     #8 Bob Snively will enter a letter ballot comment to remove annexes 
     form FC-Tape for items documented in FCP-2. Completed.
     5. Letter Ballot Comment Review:
     Editor Dave Peterson, STK, had brought a list of letter ballot 
     comments that he wanted to have the group review. The other comments 
     he felt he could handle on his own.
     #1. Bob Snively (SUN) Comment ID 47. The REC does not provide enough 
     information to create a properly formed SRR. 
     Bob Snively, SUN, referred to figure 10 in Annex B.The REC accept does 
     not provide relative offset for resumption of a transfer with the 
     Enable Modify Data Pointers (EMDP) bit set. Matt Wakeley, HP, 
     responded that this figure was just requesting the retransmission of 
     the transfer ready and would not exhibit a problem. 
     Neil Wanamaker, Crossroads, suggested that Figure 17 in Annex B was a 
     better example as it includes an error on a data transfer. The REC 
     indicates the number of bytes transferred but with the EMDP bit set 
     the sequences are not required to be sent in-order so it is not always 
     clear where the recovery should begin. Bill Martin, Gadzooks, proposed 
     that on a read failure the SRR could request the RO at the beginning 
     of the transfer. As the problem occurs when reads operations are 
     streamed, Matt Wakeley suggested precluding streaming reads, which Bob 
     observed would seriously limits tape performance. Dale suggests 
     retransmitting the whole record again, which he clarified was the 
     entire command, typically 256K bytes. Write transfers do not have the 
     problem since the transfer ready limits the recovery to only the last 
     Two possible solutions were considered: Specify the REC and SRR more 
     completely or give the host the option that redo the entire transfer. 
     Neil Wanamaker noted that reads are less common than writes so 
     reducing the performance of the read may be acceptable. Dal Allan, 
     ENDL, noted that recovery is a high anxiety moment - reducing the 
     performance of the recovery may not be good. 
     Rob Basham, IBM, interjected a case he was concerned about which he 
     called the fixed block problem. In this case recovery becomes more 
     difficult when it starts in the middle of a physical block. Dale 
     replied that the solution is to not release the data buffer until you 
     are certain that the host got it all so that you do not have to 
     recover the data from the tape.
     The conclusion was to recover streamed reads from a relative offset of 
     zero if there is any ambiguity, meaning that the entire command is 
     re-transferred. It there is no ambiguity, recovery can begin from some 
     later position in the transfer. 
     Bob Snively, who is the editor of the FCP-2, noted that some of these 
     issues belong in the FCP-2. He would like the issues to be resolved by 
     this group and placed in annexes in the FC-Tape document. He will put 
     them in the FCP-2 and they can be removed from tape document before it 
     is published. 
     Agenda item 6.1 
     For the convenience of the editor, there was a break in the letter 
     ballot review while Bob Snively reviewed the FCP-2 status.
     Bob provided an overview of the changes in FCP-2 from FCP. The first 
     topic was mode pages. The first one Bob mentioned was the 
     Disconnect-reconnect page. The next was the Fibre Channel Logical Unit 
     Control page (18). Interesting bits on this page are the Maximum Burst 
     Size field (which defines the maximum sequence length) and EMDP bit. 
     The next mode page is the Fibre Channel Logical Unit Control page 
     which contains the EPDC which enables the precise delivery function 
     (command CRNs). The last one mentioned was the Fibre Channel Port 
     Control Page (19) which contains port configuration bits. 
     In reviewing this page, Tak Seto, Quantum, noted that these 
     definitions were inconsistent with previous documentation. Bob noted 
     that these functions had been defined in several different documents 
     and that the FCP-2 would be the final authority. 
     Pak noted that DTOLI requires that FC-AL-2 behavior be required and 
     would obsoletes FC-AL devices. He asked if FCP-2 should require FC-AL2 
     compliance. This was discussed. It was noted that many of the bits 
     required FC-AL2 to be implemented anyway. This secretary was of the 
     opinion that the discussion conclusion was that AL2 compliance was 
     required. Jeff Williams, Western Digital, and Jim Coomes, Seagate, 
     agreed that these features needed to be reviewed and discussed. Having 
     Bob collect them into the FCP-2 is providing a forum for the review to 
     Jim Coomes noted that an additional bit had been proposed for the Port 
     Control page. He called it Enable auto response and it had been 
     located in byte 4, bit 0. This bit is necessary because some 
     initiators cannot accept the response in the same latency as the data 
     on a read operation. 
     George Penoki, IBM, expressed concern that Bob Snively, The FCP-2 
     editor was making technical changes to the FCP-2 without obtaining 
     approval by submitting a formal request to the appropriate groups. Bob 
     replied that the some of the tape development will be moved into FCP-2 
     without a new proposal, as he felt this requirement was met by the 
     tape discussions. He considered this presentation to be the formal 
     proposal for several of the features discussed.
     Other changes to the FCP-2 include adding Confirmed delivery, RR_TOV, 
     an extended CDB proposal, eliminating operation associators, 
     eliminating sequences with mixed command and data as well as mixed 
     data and response. The concept of confirmation is being dropped.
     After Bob concluded the agenda was returned to agenda item 5, 
     reviewing the letter ballots comments:
     #2 Matt Wakeley (HP) Comment titled: Public behavior not documented. 
     Stewart Wyatt noted that there are several places in the profile that 
     presume private behavior that need to be corrected. The biggest area 
     of concern with public behavior was the subject of authentication 
     after a loop initialization. Currently the profile refers to the FLA 
     for public initialization and copies material from the PLDA for 
     private loop initialization.   Jeff Stai, Brocade, said he had heard 
     some hallway talk about a suggestion to generate a generic SCSI and 
     Fabric behavior. Bob Snively thought that would belong in FCP-2. 
     Stewart Wyatt felt that a common initialization process needed to be 
     defined that demonstrated the process of discovering that the fabric 
     was not present and proceeding to private loop initialization. Stewart 
     was also interested in how hosts on remote loops were informed of 
     initialization changes. Jeff said that was done through state change 
     notification. Stewart noted that this was prohibited in this profile 
     which he thought was an indication that the process had not been 
     thought through thoroughly.
     In conclusion editor Dave Peterson took an action item to provide an 
     annex covering the whole process by enlisting the help of a few 
     #3 Matt Wakeley (HP) Comment titled: The Tape Profile should not 
     prohibit establish commands. 
     Dal Allan groaned, noting that this is the same argument that has been 
     rehashed many times. Dal argued that if people would only read the 
     document definition of prohibit this would not be a problem. George 
     Penoki agreed with Dal but sided with the letter ballot comment that 
     he felt that prohibited is too strong of a term. Bob Snively noted 
     that there are two parts to compliance. All required commands must be 
     implemented and all other commands must never be required. There was 
     some discussion of compliance testing. 
     Bob reiterated his goal that he doesn't want to have to consult 
     individual manuals to write a driver. He just wants to use the 
     compliant command list. Dave Peterson felt that limiting commands 
     exceeded the scope of the profile. The goal is to leverage existing 
     SCSI solutions to Fibre Channel. Rob Basham noted that the command set 
     on his Fibre Channel and SCSI drives will be the same. Rob's drives 
     will support every required command. He sees the list as a systems 
     house issue and is irrelevant to tape drives in the short term. 
     However it may become relevant in the future.
     Dave questioned the relevance of this activity. Dal and Bob disagreed 
     that a common command set is necessary for interoperability which is 
     within the scope of the project. Dal Allan noted how difficult it is 
     to remove commands. There is substantial cost savings by reducing the 
     number of commands supported because it reduces the testing required 
     for proving compliance. Dave Peterson agreed but said this effort 
     belongs in the SSC. Rob Basham felt it was a secondary issue that 
     should not delay the profile. Matt Wakeley felt that this issue is not 
     transport specific and should apply to any transport therefore it is 
     not relevant to this group. Dal agreed that Matt would be correct if 
     this is a standard, but this is an interoperability profile so it is 
     an appropriate activity. 
     Rob Basham proposed the profile will only contain commands and 
     parameters that are required for compliance to the profile. This 
     proposal passed on a letter ballot vote16 to 3.
     Matt Wakeley asked for a vote on a proposal to remove the tables that 
     contain the compliant commands, the tables 26 and 27, entirely. This 
     vote failed 2 to 12.
     The editor was instructed to remove all of the commands that are 
     prohibited from the table and update statements about 
     #4. Matt Wakeley (HP) Comment (untitled) on alerting an initiator to a 
     CRN sequencing problem. 
     The problem occurs if the target is unable to get the initiator to 
     respond to a missing CRN. The conclusion was after RR_TOV to 
     explicitly log out the initiator. 
     Stewart Wyatt questioned the assumption that the target would retain a 
     command received out-of-order. Dale Lafollette replied that the target 
     was expected to execute the commands in order thought they might not 
     be received in-order.
     #5 Charles Binford (LSI) CommentID 6. Class 3 and Link Control. 
     The note referred to in this comment suggests that an ACK should be 
     rejected in Class 3. A discussion followed about appropriate behavior 
     and confusion in other documents. Matt Wakeley proposed removing the 
     comment from profile and defer to FC-PH. This was accepted. 
     #6 Charles Binford (LSI) CommentID 8 
     Charles requested that the profile requesting text for clause 5.13 
     which is on sequence and exchange management. Previous versions of the 
     profile included text copied from the PLDA. The group had decided to 
     remove the text and refer to FC-PH. The group voted 9 to 0 to reject 
     the comment (Replacing text previously removed.) Note that Charles 
     Binford was not present to defend his comment.
     #7 Charles Binford (LSI) CommentID 9 LPE/LPB should be I not P. 
     Stewart Wyatt noted that a footnote promotes this feature while the 
     table prohibits it. Bob Snively and Dale LaFollette argued that 
     feature is not useful and that the footnote should be removed. Matt 
     Wakeley was concerned about loops with non-tape devices which may need 
     this feature as he felt it put hosts in a no-win situation. A host may 
     need to use LPEfx which applies to all devices on the loop and may be 
     useful in some environments. 
     The question was put to a vote: Shall initiators be allowed to 
     initiate LPE/LPB? (Changes P to A) 11 yes 2 no. A second question was 
     voted on: Should the footnote be eliminated altogether? 5 voted yes 
     and 8 voted to keep it. A third vote was should the note be changed? 
     Approved 13 to 0. The note should be modified to say: Allowed so that 
     the initiator may manipulate non-tape devices.
     #8 Robert Snively (SUN) CommentID 26. Incorrect value for FCP_TOV
     Bob argued that the only reasonable use for FCP_TOV is as a polling 
     interval for REC.The group agreed and the editor was instructed to 
     remove item (a) "time FCP level information unit reply Sequences" from 
     clause 7.6. The group also agree to change timer name to REC_TOV to 
     emphasize its function. 
     #9 Bob Snively (SUN) CommentID 24. Problem with E_D_TOV definitions. 
     Matt noted some errors in Bob's comments which Bob accepted. E_D_TOV 
     definition is flawed since it doesn't consider delays in the loop. 
     (This needs to be fixed but is a project for another group.) Matt 
     Wakeley pointed out that a fabric attached initiator is entirely 
     unaware of the presence of a public loop that a target is attached to. 
     Bob proposed that the timer be large enough that the loop issues would 
     be covered. The comment was accepted with Matt's corrections and a 
     note that this solution needs to implemented until the underlying 
     problem are addressed by another group. 
     Bob was concerned about implementing multiple E_D_TOV timers to time 
     every outstanding sequence. A long discussion followed to convince 
     #10 Bob Snively (SUN) CommentID 25 Incorrect definition of RR_TOV.
     Bob proposed setting RR_TOV to a constant value of 16 seconds. Rob 
     Basham said that RR_TOV has to be greater than E_D_TOV. Rob likes to 
     have the timers defined with respect to each other so that the order 
     in which they expire is always the same. This goal can also be 
     achieved if they are all defined by constant values. The proposed 
     value was questioned as being too short for some observed exchange 
     authentications to be completed as observed in the interoperability 
     tests. A question was raised about having multiple timers running when 
     a loop initialization occurs. Bill Martin noted that is unavoidable 
     since a fabric connected initiator will have an E_D_TOV time-out 
     running and will be unaware that the loop is initializing. A long 
     discussion followed about how the counters relate to each other. Then 
     Bob Snively lead an discussion to calculate the worst case value for 
     Stewart Wyatt pointed out that increasing RR_TOV requires a target to 
     keep more status assuming that status cannot be discarded until after 
     RR_TOV expires. Rob Basham gave an example of receiving many inquiries 
     from different initiators where most of them don't send a subsequent 
     command to confirm delivery of the previous status. Matt Wakeley 
     corrected the assumption that RR_TOV was the required time to retain 
     the status. The correct value is 3 times REC_TOV. Rob Basham told the 
     group that they needed to get used to the idea that targets could not 
     keep large amounts of data around for very long. The group decided to 
     separate the RR_TOV issue from holding status and consider only 
     exchange authentication for this discussion.
     Dave Baldwin suggests a constant of 5 minutes since he has seen much 
     longer periods than 16 seconds. Jim Coomes thought that 5 minutes was 
     a reasonable value from his experience with disk drives. Note that 
     RR_TOV is a tunable parameter in Mode page 19. Dave's proposal was 
     #11 Bob Snively (SUN) CommentID 30 Note (Clause 8.2.1) is incomplete, 
     data overlay
     The issue is how a host determines that all of the data was 
     transferred with data overlay, since it cannot rely on counting the 
     bytes transferred and comparing the sum to the FCP_DL when data 
     overlay is allowed. Bob proposed that the note be changed to state 
     that the method to determine that all of the data is transferred is 
     beyond the scope of this profile. 
     Bob lead the group on a tutorial of how the host might make this 
     determination. One approach is for the initiator to depend on the 
     target to inform it that all of the data has been transferred when 
     error recovery occurs with the EMDP bit set. Bob thought the host 
     could do it but that it would be so difficult that it would rely on 
     the target. Rob Basham brought up the issue of read with SILI bit set. 
     In this case the target cannot inform the host if the transfer length 
     is different than what was requested. Bob corrected him by pointing 
     out that the FCP-DL always applies and provides the expected length. 
     If the target transfers a different amount it reports it as a residual 
     in the FCP_RSP. In this profile the host is required to make the 
     expected data transfer exactly equal to the FCP_DL field.
     #12 Charles Binford (LSI) CommentID12 Limit data overlay to error 
     It appeared that Charles wanted to limit data overlay to error 
     recovery. Bob Snively felt that if it was allowed at all the target 
     could invoke it at its discretion. 
     6. T10 New Business:
     Note the discussion of the FCP-2 reported earlier.
     7. T11 New Business:
     8. Review New Action Items:
     The only new action item was for the editor to create an 
     authentication annex 
     9. Adjournment:
     The chairman noted that there will be no meeting in January. The 
     editor reported that he had reviewed the letter ballot comments and 
     had covered all of the comments in the meeting today that he believed 
     he needed guidance from the group on. He wanted to updated the profile 
     with the results of today's discussion and with the remaining letter 
     ballot comments. He felt like he could complete this by February's 
     meeting. A request was made for him to post the results two weeks 
     before the February meeting to allow the group time to review the new 
     draft before the meeting.
     The meeting was adjourned about 6 PM. 


* For T10 Reflector information, send a message with
* 'info t10' (no quotes) in the message body to majordomo at symbios.com

More information about the T10 mailing list