IEN # 3 Jon Postel Supercedes: None ISI Replaces: None 18 August 1977 1.4.1 INTERNET Meeting Notes 15 August 1977 Editors Remarks Here are some notes from the Internet meeting held at ISI on 15 August 1977. The notes are incomplete, and remarks attributed to individuals may be in error. Agenda Opening Remarks - Danny Cohen Meeting Objectives - Vint Cerf Current Internet Plans - Vint Cerf Issues in Internetting - Andrew Hinchley & Chris Bennett Gateway Monitoring - Bob Bressler International Standards - Vint Cerf X.25 Gateway Interface Experiment - Peter Kirstein Type of Service Issues - Danny Cohen Internet Mail Service - Steve Crocker Wrap Up - Vint Cerf Future Meetings - Vint Cerf Opening Remarks - Danny Cohen Cohen discussed the agenda and the rules of the meeting, Danny is moderator and can recognize people to speak or cut people off if necessary. Meeting Objectives - Vint Cerf Cerf presented the following meeting objectives: Review the current internet plans Raise and discuss internet research issues Compose a list of objectives of internet research Prepare a prioritized list of internet tasks Prepare a prioritized list of internet experiments Section 1.4.1 [page 1] 18 Aug 77 IEN # 3 INTERNET Meeting Notes Current Internet Plans - Vint Cerf Cerf reviewed the current (FY 78) Internet Plan: TCP Development UCLA - 360/91 MIT - Multics BBN - Tenex, Tops 20, Unix SRI - LSI-11 NDRE - NORD-10 UCL - PDP-9 (?) Network Development BBN - ARPANET (gateway routing, flow control, broadcasting) MIT/LCS - LCSNET, CHAOSNET Collins & BBN - Packet Radio Network Broadcasting MIT/EE - Theory (spanning trees...) Gateways BBN - Packet Radio / ARPANET BBN - Packet Satellite / ARPANET BBN - ARPANET / RCCNET BBN - Gateway Control Center XEROX - Packet Radio / Ethernet MIT - ARPANET / LCSNET(s) UCL - EPSS / ARPANET UCL - EPSS / Packet Satellite UCL - X.25 Net Issues BBN - Gateway flow control, routing, reporting, checks and balances, etc. UCLA - Internet performance modelling and analysis, gateway flow and congestion control MIT/EE - Relibility of internetworking, alternate routing ISI - Fast network deployment, addressing, authentication Section 1.4.1 [page 2] 18 Aug 77 IEN # 3 INTERNET Meeting Notes Internet Performance Measurement UCLA - Internet Measurement Center UCL - Internet Gateway traffic generation and performance measurement Internet Services ISI - Internet Mail Protocol Development ISI - Internet Types of Service (TOS) ISI - Conferencing, multi-connection synchronization LL - Packet Satellite / internet speech [NDRE, UCL, LL, BBN, LPCMs] SRI - Impact of TOS on Packet Radio ISI - Protocol Modelling and Testing/Evaluation System Cerf: We need a specification for a gateway so that other inplementations will come out the same. Cerf reviews related working groups: TCP working group ( V. Cerf - ARPA) focuses on TCP as an end to end protocol in an internet environment, technical orentation. Packet Radio working group (D. Neilson - SRI) Packet Satellite working group (I. Jacobs - Linkabit) Internet working group (D. Cohen - ISI) focuses on network interconnection problems and plans internetwork experiments. Section 1.4.1 [page 3] 18 Aug 77 IEN # 3 INTERNET Meeting Notes Cerf reviews the "number czars" for various note series etc: PSPWNs - Lin-Nan Lee (Linkabit) [JACOBS@ISIE] PRTNs - Ron Kunzelman (SRI) [PRSETD@ISIC] IENs - Jon Postel (ISI) [POSTEL@ISIB] Network Identifiers - Jon Postel (ISI) [POSTEL@ISIB] Internet Messages Types - Jon Postel (ISI) [POSTEL@ISIB] Cerf reviews the forecast of events in interneting: TCP Development Sep 77 - Experimental small-machine Top20 TCP Oct 77 - Unix TCP Nov 77 - Operational small-machine Tops20 TCP Dec 77 - Nord 10 TCP Jan 78 - Tenex TCP May 78 - large-machine Tops20 TCP Jul 78 - 360/91 TCP Network Development Sep 77 - 2 node PR test cell in Boston Jan 78 - LCS Net Jun 78 - PR Net Broadcast Dec 78 - ARPANET Broadcast Gateways Oct 77 - Ethernet/PR net Dec 77 - Gateway monitoring center Jan 78 - LCSNet/ARPANET Jan 78 - X.25 Epss interface / SATNET Apr 78 - Gateway flowcontrol, routing etc specification Oct 78 - Gateway flow control, routing, etc. Pass 1 implemented Jan 79 - PR Net in Army C2 test bed Section 1.4.1 [page 4] 18 Aug 77 IEN # 3 INTERNET Meeting Notes Performance Measurements May 78 - SATNET/ARPANET/EPSS/PRNET performance tests (UCL) Oct 78 - Internet Measurement Center (UCLA) Jan 79 - Internet performance tests Mar 79 - Rapid deployment tests Protocol Development Sep 77 - TCP 2.5 specification Jan 78 - Unix telnet Mar 78 - Tenex & Tops20 Telnet Apr 78 - Draft TOS Specification May 78 - Prototype FAX specification from UCL Jul 78 - Prototype internet mail service Aug 78 - Tenex & Tops20 FTP Sep 78 - Internet speech conferencing specification Oct 78 - 360/91 telnet & ftp Mar 79 - Internet speech conferencing tests Jun 79 - Speech & Video conferencing specification Demos Sep 77 - Point to point raw internet speech demo May 79 - Internet services demo Cerf: We need a specification of how a gateway should behave to cooperate with a gateway control center. Kirstein: EPSS and SATNET will have X.25 interfaces but there is no plan for an X.25 interface to the ARPANET. Cerf: The issue of secure internetting is postponed. Cerf: There is consideration of an effort to develop a new ftp, but there is no plan as yet. Kirstein: FAX might be worked into the speech and video conferencing system. Section 1.4.1 [page 5] 18 Aug 77 IEN # 3 INTERNET Meeting Notes Kirstein: Are LLL or Commercial nets to be involved in internet experiments? Cerf: LLL is interested in TCP for the local to Livermore Octopus net not the CTR net. Cerf: No current plans to get commerical nets involved. Kirstein: What is happening about AUTODIN II and ARPANET as described at TCP Meeting? Cerf: That plan is no longer active, and it is not expected that there will be any move to get people off the ARPANET until 1980, after that AUTODIN II service may be offered instead of ARPANET on a customer by customer basis. Kirstein: It seems more likely that university type users will have to go to a commerical net than get on AUTODIN II, so shouldn't we be exploring internetting with commercial nets now? Issues in Internetting - Andrew Hinchley & Chris Bennett Hinchley & Bennett discussed the paper (PSPWN 76, INDRA 637) "Issues in the Interconnection of Datagram Networks" that was distributed prior to the meeting. Hinchley: there are three catagories of interconnections: 1) ARPA like datagram networks, 2) PTTs X.25 networks, note that the PTTs are moving very fast with a model of an internetwork system composed of a small number of very large national networks, 3) Interconnection of private networks via public X.25 networks. The SATNET experience is the basis for these (PSPWN 76) comments. Cerf: Is a gateway a piece of hardware physically connected to two or Section 1.4.1 [page 6] 18 Aug 77 IEN # 3 INTERNET Meeting Notes more networks, or is it pair(s) of software modules connected by physically complicated networks ? Kirstein: We should be more aware of what is going to be possible with PTT provided "virtual call networks". Bennett: Addressing is a topic discussed in the memo. Cerf: Up to now we have not had names for gateways, but now we may need names if gateway control centers are to communicate directly with gateways. Cerf and Crocker discuss gateway models, one point was that one should try a model with a three or more connected gateways since sometimes a two connected model may make things seem simpler than they really are, another point is that a gateway is a kind of a forwarder, and Crocker argues that the forwarder is logically a network. Postel: An address is just a string of bits, with each address parser consuming part of the string, to give a gateway an address reserve one value of the set of values representable in the bits the gateway parses, that reserved value means "here". Kirstein expressed concern about the address space in the individual networks. Also concerned about translations necessary when going between datagram and virtual call networks. Cerf: Does the way a thing is named specify the way one gets to it? Especially if a thing is multi-hommed? Shoch: One way is all gates are in the same net and have exactly one address. Another way is that a gateway has distinct host address on each net it is connected to. Cerf: If the gateway control center has to talk to the gateway and the gateway has two names, which name does the control center use? Shoch: Someone has to take responsibility for choosing. Jacobs: Where do the global unique names come from? Section 1.4.1 [page 7] 18 Aug 77 IEN # 3 INTERNET Meeting Notes Postel: They are external. We always have a unique description, usually by location, for objects. The gateway control center will have to have for each gateway a unique (external) name, and associated with that a list of internal addresses. Cerf designates that there be a committee to study addressing and naming of gateways to (1) figure out what things there are, and (2) figure out how to address them. Crocker is chairman. Bennett: Routing in the Internet is discussed in PSPWN 76. The main ideas are that adaptive routing in internet is not likely to be much good due to the larger delays in information propagation. Fixed routing with a number of alternatives (this is also called "Explicit Path Routing") might be useful. [Naylor says this is explored by IBM Zurich.] Could use alternate routing in a event driven form to recover from service outages. One issue is what to base the adaptation on. Postel: Most of the things normally used could get washed out by local traffic. Crocker: Perhaps there are some things... Postel: for example queue lengths in the gateways. Cerf: Adaptive routing is not a substitute for congestion control. Jacobs: The information available to a gateway includs both facts about its directly attached networks and information received from other gateways. Cerf: Gatways may have to exchange information about what paths are viable. If one packet from a gateway to a host in the ARPANET results in a "host dead" error what does the gateway do about other packets to the ARPANET? We should avoid having a lot of state information in the gateways. Postel: What happens when the ARPANET says to a gateway "host dead"? Jacobs: Nothing. Section 1.4.1 [page 8] 18 Aug 77 IEN # 3 INTERNET Meeting Notes Cerf: Try to pass an error message back to source TCP. We have no conventions for this as yet. Mathis: Out thinking has been: Gateways do routing; TCPs do not do routing. Cohen: Routing algorithms need to be smarter. Hinchley: Flow Control Possibilities 1) No control at gateway level (expect end-to-end control to take care of any problems) 2) Gatway to Gateway control 3) Terminating Gateway Control (source - desination gateways control flow) 4) X7X virtual circuit control This is a range of possibilities. Postel favors a hop by hop (that is gate by gate) flow control. Cohen: Why should there be a single type of flow control? The type of service (drops, streams, floods) should get different kinds of flow control. Jacobs: In SATNET there are a whole rage of stratagies from try once to try very hard to get a message through. There occured at this point a discussion of the cost of hop by hop acknowledgements. Jacobs: This should all be in terms of bump to bump, that is a gateway consist of a bump for each connected network and a core, any network is between a pair of bumps, and the bumps should decide how much error control etc is needed across that network. There was a discussion of status information, particularly error Section 1.4.1 [page 9] 18 Aug 77 IEN # 3 INTERNET Meeting Notes reports. If a gateway has a problem then it is an internet error and "the internet" ought to tell the source about it. If a destination host has a problem then a service center could know the problem and a query to the services center could tell the source what is happening. Hinchley: Gateway Control, Once gateways get above a certain level of complexity there needs to be a control function, that suggest that the reshould be a gateway control language which is standard throught the internet. Gateway Monitoring - Bob Bressler Bressler described the current gateway monitoring center as a fairly simple program that summarized reports from the SIMP and some other gateway. The information reported was: number of messages sent and received on each port of each gateway and the number of messages dropped by the gateway The gateways spontaneously generate the reports to the control center, the control center curently sends no messages to the gateways. There are data generators and data sinks in the gateways now controled by XNET a ddt like cross net debugger. Kirstein: One of the simplest services one could provide is reliable delivery but allow duplicates to be delivered as well. Hinchley: What common functions can be listed? restart reload run a diagnostic update routing information Shoch: Gateways exchange routing tables perodically gratuitously. Many processes in a gateway for example time server, name lookup server, boot loaders, measurement process, echo server, source, sink, trace. Shoch: There is a distinction between Routing, Addressing, and Naming. Section 1.4.1 [page 10] 18 Aug 77 IEN # 3 INTERNET Meeting Notes Kirstein: There will need to be several monitoring centers. There may be a central control center that forwards suggestion for control to national control centers. Burchfiel: No gateway in the ARPANET currently does fragmentation. International Standards - Vint Cerf Cerf reviews international standards organizations: ITU - International Telecommunications Union (established by treaty) CCIR - International Radio Consultative Committee (issues recommendations on the use of electromagnetic spectrum) CCITT - International Telegraph and Telephone Consultative Committee (issues recommendations on telephone and telegraph communication) PTTs - Government operated Post, Telephone & Telegraph Systems ATT - American Telephone and Telegraph Corporation IFIP - International Federation for Information Processing ANSI - American National Standards Institute cats & dogs common carriers manufacturers EIA - Electronic Industry Association government NBS - National Bureau of Standards DCA - Defense Communications Agency ISO - International Standards Organization ANSI PTTs manufacturers IFIP CEPT - UK France DBP Section 1.4.1 [page 11] 18 Aug 77 IEN # 3 INTERNET Meeting Notes ECMA - European Computer Manufacturers Association CII HB Phillips ICL Cerf reviews relevent standards activity: Analog V.24 (CCITT recomendation) = RS232C (EIA) Digital X.21 <-> RS232C HDLC (ISO) <-> [SDLC (IBM), UDLC (Univac), ADCCP (ANSI)] X.25 packet network interface, virtual call, asymetric X.C PAD control X.7X packet net / packet net interface (virtual circuit) X.25 Gateway Interface Experiment - Peter Kirstein Kirstein discussed the work UCL is going to be doing with EPSS and X.25 interfaces. Section 1.4.1 [page 12] 18 Aug 77 IEN # 3 INTERNET Meeting Notes Type of Service Issues - Danny Cohen Cohen discussed Type of Service issues. He suggests the following dimensions to service selection: priority relibility/cost delay/cost Floods / Streams / Drops Message/Wire Sort Messages Acknowledgement / Negative acknowledgement/ No acknowledgement Security Synchronization Multidestination addressing End of Letter Buffer Runout Receiver Wakeup Cohen: Whether ACK or NACK or both or neither are used should depend on statistics. Jacobs: The Host-SIMP protocol paper may be relevant. Hoversten: Is the user helping the network by telling it what type of demands the user is going to make? Internet Mail Service - Steve Crocker Crocker discussed a model of how to make mail services available in the internet system in a straightforward way. Also of interest is possibility for authentication and privacy. Section 1.4.1 [page 13] 18 Aug 77 IEN # 3 INTERNET Meeting Notes Wrap Up - Vint Cerf Binders for an Internet Notebook were distributed, and Cerf explaind the intention to use it to collect the current information for internet experements. Jon Postel is the Notebook coordinator. Cerf summarized as follows: Addressing alternatives need to be written down - assigned Steve Crocker to coordinate. Please supply him with your thoughts. What objects exist and what names should they have. John Shoch will provide a one page note on nomenclature. A Gateway Specification is to be prepared by Bob Bressler. Cerf will distribute a current version of the X.25 specification. Kirstein will distribute a document on the UCL X.25 experiments. Future Meetings - Vint Cerf Cerf reviewed the plans for future meetings: 15 Aug 77 - Internet meeting at ISI 17-19 Aug 77 - PSPWG meeting at Linkabit 13-14 Oct 77 - TCP meeting at SRI 31Oct-2Nov77 - PSPWG meeting at BBN 3 Nov 77 - Internet meeting at BBN 30-31 Jan 78 - TCP meeting at ISI 1- 2 Feb 78 - PSPWG meeting at UCLA 3 Feb 78 - Internet meeting at UCLA 20-21 Apr 78 - TCP meeting at BBN 1- 2 May 78 - Internet meeting at UCL 3- 5 May 78 - PSPWG meeting at UCL 13-14 Jul 78 - TCP meeting at PARC 2- 3 Aug 78 - Internet meeting at LL 4 Aug 78 - PSPWG meeting at LL 12-13 Oct 78 - TCP meeting at LCS 2- 3 Nov 78 - Internet meeting at SRI Section 1.4.1 [page 14] 18 Aug 77 IEN # 3 INTERNET Meeting Notes Attendees: Chris Bennett UCL UKSAT@ISIE (01)387-7059 Bob Bressler BBN BRESSLER@BBN (617)491-1850 Jerry Burchfiel BBN BURCHFIEL@BBNA (617)491-1850 Vint Cerf ARPA CERF@ISI (202)694-3049 Danny Cohen ISI COHEN@ISIB (213)822-1511 Steve Crocker ISI CROCKER@ISI (213)822-1511 Jin Forgie MIT-LL FORGIE@BBN (617)862-5500x7191 Andrew Hinchley UCL UKSAT@ISIE (01)387-7050x817 Estil Hoversten COMSAT HOVERSTEN@ISI (202)554-6092 Irwin Mark Jacobs LINKABIT JACOBS@ISIE (714)453-7007 Peter Kirstein UCL KIRSTEIN@ISI (01)387-7050 Ron Kunzelman SRI KUNZELMAN@ISIC (415)326-6200x4118 Jim Mathis SRI MATHIS@SRI-KL (415)326-6200x5150 Bill Naylor UCLA NAYLOR@ISI (213)825-4864 Jon Postel ISI POSTEL@ISIB (213)822-1511 John Shoch XEROX-PARC SHOCH@PARC (415)494-4000 Paal Spilling NDRE PAAL@SRI-KA Dan Wilhelm SAMSO SAMSO@ISI (213)643-0131 Section 1.4.1 [page 15]