IEEE P1355 Working Group Second working group meeting 22nd June 1993 Minutes The second meeting of the IEEE P1355 (formerly known as IEEE HIC) Working Group was held at Bull SA, Brussels, 11.00 - 17.00 on 22nd June 1993 Present Ralf Bok„mper Harting ElektroniK GmbH Stefan Haas CERN Ernst Kristiansen SINTEF Roland Marbot Bull SA Paulo Melloni Bull SpA Colin Whitby-Strevens INMOS Limited Apologies Andrew Cofler Bull SA Peter Thompson INMOS Limited Brian Martin CERN Sylvain Paineau THOMSON RCM 1 Welcome, introductions, approval of agenda CWS welcomed everyone to the meeting, thanking Bull for their hospitality. CWS explained that the IEEE meetings were run according to Roberts Rules, and that therefore the first action is to approve the agenda. The agenda was approved, subject to re-ordering so that all non-technical items were dealt with first. 2 Approval of minutes of 24th February meeting The minutes, previously distributed, were approved as a correct record. 3 Chairman's report 3.1 PAR Approval CWS reported that the sponsorship of the PAR had been discussed at the IEEE Computer Society Standards Activities Board, where the Microprocessor Standards Committee had expressed the view that the PAR fell within their scope. As an interim measure, and to ensure no hold-up to the progress of the PAR, the SAB had decided to sponsor the PAR directly, until such time as it is decided which of BASC (the original sponsor) and MSC should be the permanent sponsor. Meanwhile, CWS reported that he is sending the status reports to all three chairmen. The PAR was duly discussed at the meeting of NESCOM on Tuesday 16th March. NESCOM had asked that the IEEE requirements on licensing be duly observed, and had asked for additional coordination. It then approved the PAR. The PAR was then placed, with others from NESCOM, on the consent agenda for the IEEE Standards Board (17th March), where it was approved without discussion. Prior to NESCOM, the IEEE secretariat allocated the number P1355 to the PAR. CWS reported that the ATM Platform was approved at the same series of meetings, and had been allocated the numbers P1356 and P1357. 3.2 Press Release CWS reported that he has prepared a Press Release, announcing the creation of the WG. He had been hoping to release this in order to advertise today's meeting, but this had proved impracticable. After discussion, it was agreed to aim to make the release in mid-July (at a precise date to be determined by the PR staff in INMOS and Bull SA), with a possible follow-up after the October BASC meeting. Press contacts in EE Times and Electronique International were identified. RB indicated that Harting would like to make the same release at the same time, EK stated that he would like to release through SINTEF to Norwegian sources, and would also liaise with Dolphin. 4 Review of action items from previous meeting 4.1 Reflector. The reflector is now thought to work. However, there have been no new documents to put on it. 4.2 Release of optical interconnect results. CWS reported that he had discussed the release of optical interconnect results with the manager of the GP MIMD project, who had also discussed this with CERN. SH confirmed that it had been agreed to release the interim results, and he tabled two substantial documents (totalling 70 or 80 pages, with many diagrams, illustrations and tables), entitled `T9000 Fiber Optic Link Extensions; Report on 850nm Fiber Optic Transceiver', and `Long Distance Differential Transmission of DS Links over Copper Cable'. These were copied and distributed to those attending. CWS agreed to hold master copies of the documents, and copy and distribute them on request. SH agreed to provide CWS with compressed postscript versions to be placed on the reflector. (Note in minute: these documents will form IEEE P1355 Working Papers Nos 1 and 2 respectively, for identification purposes.) CERN were thanked for releasing these results. 4.3 DS-Link connector requirements. CWS and RB reported that neither had received any feedback. The action remains open. 4.4 Conformance subsets. CWS reported that he had received no views. 7 Future meetings (item taken out of order) PM confirmed the invitation for the next WG meeting to be held in Milan on August 31st. The following meeting is planned to be held in conjunction with BASC in Boston (week of 18th October). CWS took an action to request a full day not clashing with the ATM Platform meetings (probably 19th October). It was suggested that there should be a half day tutorial on Monday 18th October in the afternoon. 5 Review of draft standard presentation. CWS reported that AC had not been able to prepared the initial draft in time for the meeting. He tabled a document which comprised a collation of the information so far `prepared', or at least identified as providing relevant technical content. He also indicated that the current plan is for AC to spend a week during the second half of July at INMOS, preparing a proper initial draft. The meeting then proceeded to consider the information in the `initial collation'. The following significant issues were discussed:- Contents Section 5.1 (Copper interconnect) should be divided into two sections, one dealing with on-board electrical properties, one dealing with off-board electrical properties, connectors and cables. References References need to be made to ATM, ATM Platform, SCI, Fibre Channel, (any other similar?), and DRAFT IEC 48B(SEC)229. Section 4 Protocol Stack This should be enhanced with the stack diagram. Level 4 should define the concept of a transaction level, but not define any actual protocols at this level. An informative annex should provide guidelines on implementing transaction level protocols, for example the constraints in order to be compatible, some example mappings (eg for ATM, Fibre Channel, . . .), etc. 5.1.1 DS-Link (copper) connector. RB indicated that he would be providing revised diagrams, and agreed to investigate their availability in machine readable form (postscript or `source'). RB presented a table specifying various aspects of the connector. This was reviewed in detail. It was agreed that all the aspects are relevant to the standard, and that the table should be included, with the deletion of the reference to the Madison cable (see below). In addition, the pin numbering for the pins should be included, as well as the identification of the lead pin (ground). The assignment of the DS-Link signals to pins should be specified (information from Harting or INMOS). 5.1.2 DS-Link Cable RB indicated that a specification for the Madison cable existed. This should be inserted into the standard. The cable spec should specify the `twists' so that the correct input at one end is connected to the correct output at the other end. 5.1.3 DS-Link Signal This should be split into two sections. One for the signals to be used for off-board connections via the connectors and cables. It was agreed that this should be pseudo-ECL differential drivers (see specifications in Working Paper 2). The other is the section in the collation on pin to pin specifications (in principle on- board, but could also be through backplane connectors - see below). The initial draft was discussed at length. It was agreed that this is too much like a data-sheet, and not enough like a standard, and a range of changes was discussed. It was proposed that the two tables (for output drivers and input pads) should be combined, as the standard should specify the electrical quality of the interconnect, not the way it is driven and received. Suggested changes included specifying a range for the operating data rate (eg 1K bit/sec - 100 Mbit/sec), specifying tph and tpl for 10 ns time period (for 100 Mbit/sec), specifying the best possible output impedance, with a tolerance (eg 100 ohms, +- x%), replacing leakage current with a min parallel resistance specification. CWS agreed to discuss the suggested changes with technical experts at INMOS and prepare a revised draft. He also indicated that a version would be prepared for 200 Mbit/sec. Signal level Minor changes required. - Should there be a skew spec? (up to 1 bit time?) 5.2 DS-Link Optical interconnect CWS reported that Sylvain Paineau had apologised for not having prepared the information in time for the meeting (partly as a result of the hand-over from Bruno Houssay, who is now in Japan), but that the information will be available shortly (in time for the mid-July drafting exercise). 5.2.1 DS-Link Optical connector An open issue is whether there should be standard single connector for each fibre, or whether there should be arrays in one connector. It was suggested that there should be the same `standard' for the optic connector as for the copper connector - ie one fibre in each direction in a single connector. It was also strongly felt that there should be only one standard (unlike Fibre Channel). A major concern is to minimise the physical space taken up by optical connectors - with the technology being able to integrate up to 32 links on a chip, and the ability to build routing switches with valencies of 1000's of links on a single board, the physical space for the connectors becomes the limiting factor to the level of integration. 5.3 Character encoding No issues 5.4 Exchange level It was queried whether this can be the same for DS and HS links. RM agreed to investigate whether there is sufficient commonality to construct a combined section. However, there are differences in the detection of disconnection, reset and start-up specifications, as well as the obvious difference of different sized flits. Item 6 was taken at this point in the meeting, as CWS had shortly to leave. 6 Review of open issues The following open issues (references are to the minute number in the minutes of 24th February) were discussed, and resolutions proposed as indicated:- 8.2 Conformance subsets. It was agreed that the only conformance subsets should be those defined by complete `branches' of the protocol tree - from `electrical' level to packet level. There would be as many conformance subsets as there are `electrical' levels. 8.4 Other standards - it was agreed that it is out of scope to define mappings of other standards, or to define bridges, etc. However these would be covered in informative annexes which would give some examples of use. 8.6 Backplane - It was agreed that it is out of scope to define any specific properties associated with backplanes. Any implementation supporting chip-to-chip connection through a backplane would simply have to satisfy the chip-to-chip electrical requirements. 8.7 Conformance testing. It was agreed that there should be an informative annex giving suggested test methods for electrical level specs. 8.8 CSRs, run-time control, interrogation. Given that the purpose of any application is out of scope of the standard, for example there might or might not be a microprocessor in the application, it was agreed that it is out of scope to specify any particular method for achieving these functions. However, it was suggested that it would be useful to have some examples of use in an informative annex. At this point, CWS handed the chair of the meeting to RM. 5 Review of draft specification (continued) Sections 6.1.1, 6.1.2. HS link Copper connector and cable. The `initial collation' has nothing on connectors or cables. RM agreed to prepare first draft tables, by adapting the tables prepared for DS-Links, and send them to RB for review, and forwarding for inclusion in the first draft. RB indicated that Harting have developed a mini-coaxial connector for differential transmission, on a pitch of 2.5mm. RB will send specifications and prototypes to RM for characterisation. Section 6.1.3. HS link Copper Signal. RM will add specifications on HS links in terms of performance, and the low frequency cut-off for AC coupling (in order to maintain DC balance). Section 6.2. HS link optical interconnection. Nothing has been done for optical connection or cables for HS links. RM will request Sylvain Paineau to consider the contents of an initial proposal for this section. Section 6.3 HS link character encoding. RM described the character encoding issues which have resulted in the design of the 8B12B encoding. He pointed out that this encoding provides: a) clock recovery using relatively low frequency (100 MHz) on-chip clocks - thus permitting CMOS implementation; b) DC balance; c) Out-of band characters for control purposes (at the exchange and packet levels, with others available for transaction layer protocols); d) per-character parity, to support error detection capabilities. Section 6.4 HS link exchange. RM described the flow control mechanism used at the sub- packet level. Section 7 Packet common layer It was agreed that routing protocols should not be included in the standard. The standard defines a packet format (with header, etc). However, the interpretation of the header is a matter for the implementation. An implementation of a routing switch would define the numbering scheme. The headers could be interpreted as node addresses, or memory addresses, virtual channels, etc. An informative annex would describe example routing schemes. It was questioned whether it will be necessary to place restrictions on packet length. It was agreed that there should not be any restriction, to allow IEEE P1355 conformant implementations to support higher level protocols efficiently (eg ATM with 53 byte cells, Fibre Channel with 1024 byte packets, SCI with three defined packet sizes, . . .). An informative annex could describe the trade-offs of directly mapping higher level transactions 1-1 onto packets, as against mapping 1-many (ie many IEEE P1355 packets to support a single higher level packet - which would allow `fairer' sharing of common links etc). The error detection and correction policy needs to be defined - and is considered an open issue particularly for HS links, as it has not been possible yet to fully evaluate BERs for these links. 8 Open issues The following list of open issues was identified:- - optics (DS and HS) - packet level (mainly HS) - conformance - DS: differential signals - commonalities in the exchange layer between DS and HS - HS: signals, packaging, connectors, cables - error detection and correction policy It was agreed that a timetable should be drawn up for resolution of these open issues before the next meeting. ACTION CWS. C Whitby-Strevens - Chairman R Marbot - Co Chairman 23rd June 1993