Stevens@A.ISI.EDU (Jim Stevens) (12/19/87)
DISCUSSION OF LAYERED PROTOCOL MODEL Jim Stevens (Stevens@A.ISI.EDU) 18 December 1987 INTRODUCTION Recently CMC has distributed a nice multi-colored wall chart showing the DoD Internet Architecture and the protocols at each layer. I only list CMC as reference because their chart caused a flurry of e-mail messages on TCP-IP. Note that their chart is just the latest of many papers and reference-type-material to contain the same type of problems. Since many of you may not have seen the CMC chart, it is included for reference as Figure 1 at the end of this message. There are 2 problems with this chart: (1) Location of EGP, GGP, and HMP and (2) Incorrect Physical and Data Link protocols under some networks (especially radio networks). LOCATION OF NETWORK MANAGEMENT PROTOCOLS The location of EGP, GGP, and HMP have caused a flurry of e-mail messages on TCP-IP. These 3 protocols are network management protocols. There has been a lot of work in the last decade on protocol architectures which include network management that provide an answer to the question of the network management protocol location. Some of the efforts include ISO (reference ISO 7498 Part 4), CCITT (ISDN, see recommendation I.320), and AT&T (J.W. Timko, AT&T Technology, Vol 2, No 3). I intend to locate EGP, GGP, HMP, and ICMP within the ISDN protocol stack using the OSI management structure categories. All three of the above named efforts basically came up with the same architecture. These architectures just added a new dimension to the current the 7-layer OSI reference model. The 7-layer OSI model contains only 1 dimension, which I'll call the X dimension. The X dimension described the different services performed upon User data as it is sent from one host to another. The new architectures added a Y dimension which showed the different functions which must be performed at each layer. The three different functions that are performed are: 1. The (N)-layer performs some service upon user data and send the data onward. (This function is identical to that described in the original X dimension.) 2. The (N)-layer performs signaling between the different layers. (For packet switched networks this signaling is typically provided in the packet headers.) 3. The (N)-layer performs internal management decisions. Figure 2 shows the ISDN protocol stack and the 2 dimensions of their model. The user service function is the vertical stack of layers labeled with a U. The signaling function is the vertical stack of layers labeled with a C. The management function is the vertical stack of layers labeled with an M. The OSI management framework describes 3 management structure categories: 1. Systems Management System Management Application-Process (SMAP) System Management Application-Entity (SMAE) 2. (N)-layer management coordinates multiple instances of communications 3. (N)-layer operation coordinates single instance of communications The Systems Management is an application process which can use the entire protocol stack when communicating. The difference between a Systems Management application process and another user application process (such as Terminal Emulation) is that the Systems Management application has access to the control/management data within each of the protocol layers. EGP, GGP, and HMP are system management applications. The (N)-layer management/operation are processes within the (N)-layer. Thus these processes can only use the protocols which lie under them. ICMP is a Network-layer management process. INCORRECT PHYSICAL AND DATA LINK PROTOCOLS UNDER SOME NETWORKS The CMC chart has confused the Physical and Data Link protocols used within some networks and the Physical and Data Link protocols used to interface to these networks. This has been a common problem and has shown up in several different papers. Figure 3 shows a layered protocol chart which did not have this problem. Since I have worked on the DARPA Packet Radio Program and its follow on Survivable Adaptive Networks (SURAN) Program, I will use older fielded Improved Packet Radio (IPR generation Packet Radios) as an example to explain how the confusion arises. Normally, the 7 protocol layers are implemented in a single box. The protocol drivers are then contained within either the OS kernel or applications residing above the OS. This is not true however in Packet Radio, where the 7 protocol layers are implemented in 3 different boxes interconnected via a BBN 1822 line. Figure 4 shows how the DOD protocols stack is implemented in 3 boxes interconnected together with BBN 1822. The Hosts interface to the Packet Radio Network via a Host Interface Unit (HIU). The actual interface between the HIU and the Hosts is a defined host access protocol running on top of BBN 1822. The hosts implements the Session and higher layer protocols such as FTP and mail. The HIU interface to the actual Packet Radios (PRs) using a defined device access protocol running on top of BBN 1822. The HIUs implement the Internet and Transport protocols such as IP and TCP. The PRs communicate to each other using the PR Channel Access Protocol (CAP) at the network and upper data link layers, CSMA at the lower data link layer, and spread spectrum at the physical layer. The PRs thus implement the Network, Link, and Physical layers. There is often confusion between the host or device interface to Packet Radio and the actual protocols used within Packet Radio. That is why the CMC chart shows BBN 1822 as the Packet Radio Network Physical protocol instead of spread spectrum. It is fairly easy to catch this type of error for radio networks, since the radio network physical layer must be some sort of RF transmission scheme such as Spread Spectrum (SS) or Amplitude Modulation (AM). ------------------------------------------------------------------------------ Figure 1 Layered Protocol Model Taken from CMC chart, Fall 1987 DOD MODEL +-------+ +------+ +-------+ +--------------------------+ +-------+ Appli- | File | | Mail | | Term- | | User Programs | | Name | cation | Trans-| | Text | | inal | | | | Server| | fer | | RFC | | Emul- | | | | | | Server| | 822 | | ation | | | | | +-------+ +------+ +-------+ +--------------------------+ +-------+ ^ ^ ^ ^ ^ ^ ^ . . . . . . . . . . . . . . v v v v v v v +-------+ +------+ +-------+ +-----------+ +------+ +-----+ +-------+ Util- | FTP | | SMTP | |Telnet | | NETBIOS | | TFTP | | NFS | | NSP | ity | MIL | | MIL- | | MIL- | | RFC | | | | | | | | STD | | STD | | STD | | 1001/ | | | | | | RFC | | 1780 | | 1781 | | 1782 | | 1002 | | | | | | 882 | +-------+ +------+ +-------+ +-----------+ +------+ +-----+ +-------+ ^ ^ ^ ^ ^ ^ ^ ^ . . . . . . . . .............................. ............................ . . v v +-----+ +-----+ +-----+ +------------------+ +-------+ Trans- | EGP | | GGP | | HMP | | TCP | | UDP | port | | | | | | | MIL-STD 1778 | | RFC | | | | | | | | | | 768 | +-----+ +-----+ +-----+ +------------------+ +-------+ ^ ^ ^ ^ ^ . . . . . . . . . . .................................................... . v Inter- +------------------------------------+----------------+ net- | Internet Protocol (IP) | ICMP | work | MIL-STD 1777 | RFC 792 | +------------------------------------+----------------+ ^ . . .............................................. . . . . . . . . . . v v v v v +-----+ +-------+ +---------+ +------------+ +--------------+ Net- | Pkt | | BBN | | Pkt | | CCITT X.25 | | Ethernet ARP | Work |Radio| | 1822 | |Satellite| |Packet Layer| | RFC 826 | +-----+ +-------+ +---------+ +------------+ +--------------+ ^ ^ ^ ^ ^ ^ . . . . . . . . . . . . ....... ......... ....... . . . . . . . . . . . . . . v v v v v . +------+ +-----+ +-----+ +-----+ +--------------+ +-------+ Data . | BBN | | BBN | |LAPB | |LAPB | |IP / Ethernet | | IEEE | Link . | HDH | | VDH | | BSC | |HDLC | | RFC 894 | | 802.2 . | | | | | | | | +--------------+ +-------+ . | | | | | | | | ^ ^ . | | | | | | | | . . . | | | | | | | | . ....................... . | | | | | | | | . . . . . . | | | | | | | | v v v v v . | | | | | | | | +------+ +-----+ +-----+ +----+ . +------+ +-----+ +-----+ +-----+ | | | | | | | | . ^ ^ ^ ^ | | | | | | | | . . . . . | | | | | | | | . .......................... | | | | | | | | . . . . . | | | | | | | | v v v v v | | | | | | | | +----+ +-----+ +-----+ +------+ +------+ | | | | | | | | Phys- |BBN | |CCITT| | EIA | | EIA | |MilStd| | IEEE | |IEEE | |IEEE | |FDDI| ical |1822| |V.35 | |RS449| |RS232C| | 188C | |802.3 | |802.4| |802.5| | | +----+ +-----+ +-----+ +------+ +------+ +------+ +-----+ +-----+ +----+ ------------------------------------------------------------------------------ FIGURE 2. ISDN Protocol Stack Taken from CCITT I.320 Control User Data System Management ________ ________ ________ | \ \ | \ \ | \ \ | \ \ | \ \ | \ \ \ \ ______\ \ \ ______\ \ \ ______\ \ | | \ | | o \ | | \|______| \|______| o \|______| @ @ o o @ @ @ o o @ @ oooooo@ooooooooo o @ @ o @ o @ @ o @ ooooooooooo @ ______@____ _o_____@_ o @ | \ @ \o @ \o @@@@@@@@@ | \ C o \ @ o \ @ | \ ________\@ o \ @ | \ |\ @ \ o M @\ | \ | \ U o\ @ \ | \ | \ _________\_________\ | \ |\ | | | | \ | \ | 7 - Appl | | | \ | \|__________| | | \ |\ | | | | \ | \ | 6 - Pres | | | \ | \|__________| | | \ |\ | | | | \ | \ | 5 - Sess | | | \ | \|__________| | | \ |\ | | | | \ | \ | 4 - Tran | | | \ | \|__________| | | \ |\ | | | | \ | \ | 3 - Netw | | ______________| \ | \|__________| | . \ |\ | | | . \ | \ | 2 - Link | | . . \ | \|__________| | . . \ | | | . . \ | 1 - Phys | | . . _______________\|__________|_________| . Physical Media | . ___________________________| ------------------------------------------------------------------------------ Figure 3. Layered DOD Reference Model Chart Taken from "Tactical Information Exchange (TIE) Framework Development (For the 1990s Time Frame" October 1981 +-----+ +----------+ +------------+ Application | MTP | |Multimedia| | Conference | | | |Message | | | +-----+ +----------+ +------------+ ^ ^ ^ ^ . . . . . . . . . . . . +------+ +---+ . . +----+ +------+ . . Utility |TELNET| |FTP| . . |TFTP| |Name | . . | | | | . . | | |Server| . . +------+ +---+ . . +----+ +------+ . . ^ ^ . . ^ ^ . . . . . . . . . . . . . . . . . . ................. ....... . . . . . . . . . . v v v . +------------------+ +---------+ +-----+ +--------+ . Transport | TCP | | UDP | | GGP | | NVP II | . +------------------+ +---------+ +-----+ +--------+ . ^ ^ ^ ^ . . . . . . . . . . . ................................ ......... . . . . v v +-----------------------------------------------+ - - - - - - - - - Internet | Internet Protocol | Stream Protocols ! +-----------------------------------------------+ - - - - - - - - - ^ . . ...................................................... . . . . . . . . . . . . v v v v v v +----------+ +------+ +------+ +-----+ +-----+ +------+ Network | ARPANET | |SATNET| |CCITT | |PRNET| |JTIDS| | DIN | | HOST/IMP | | | | X.25 | | | | | |2 SIP | +----------+ +------+ +------+ +-----+ +-----+ +------+ ^ ^ ^ ^ ^ ^ . . . . . . . . . . . . .............. ......... . . . . . . . . . . . . . . . . . . . . . . v v . v v v v v v +----+ +----+ . +-----+ +----+ +-----+ +-----+ +-----+ +------+ Link |HDH | |VDH | . | VDH | |HDH | |HDLC | |CSMA | |TDMA | |ADCCP | |HDLC| |RTP | . | RTP | |HDLC| | | | | | | | | +----+ +----+ . +-----+ +----+ +-----+ +-----+ +-----+ +------+ ^ ^ . ^ ^ ^ ^ ^ ^ . . . . . . . . . . . . . . . . . . v v v v v v v v v +---+ +-----+ +----+ +-----+ +----+ +-----+ +-----+ +-----+ +------+ Physical |M/W| |Modem| |BBN | |Modem| |M/W | |X.21 | | SS | |SS/FH| | M/W | | | | | |1822| | | | | | | | | | | | | +---+ +-----+ +----+ +-----+ +----+ +-----+ +-----+ +-----+ +------+ Key: ADCCP Advanced Data Communications Control Procedures DIN 2 SIP AUTODIN II Segment Interface Protocol FH Frequency Hop HDH HDLC Distant host JTIDS Joint Tactical Information Distribution System MTP Mail-Transport Protocol M/W Modem / Wire NVP II Network Voice Protocol PRNET Packet Radio Network RTP Real Time Protocol SATNET (Experimental) Satellite Network SS Spread Spectrum TDMA (Synchronous) Time Division Multiple Access VDH Very Distant Host ---------------------------------------------------------------------------- Figure 4 Host Interface to Packet Radio +---------------+ | | Host Application |FTP, Mail, etc.| | | Utility | | +---------------+ ^ . BBN 1822 and defined host access protocol . v +---------------+ | | Host Transport | TCP | Interface | | Unit | | Internetwork | IP | | | +---------------+ ^ . BBN 1822 and defined device access protocol . v +---------------+ | | Packet Network | Channel | Radio | Access | | Protocol | | | | | Link | CSMA | | | | | Physical | Spread | | Spectrum | | | +---------------+ ---------------------------------------------------------------------------- -------
karn@faline.bellcore.com (Phil R. Karn) (12/19/87)
I think the reason there is so much argument about where things like network management protocols and packet radio protocols "go" in the ISO/OSI reference model is that the OSI reference model itself is fundamentally flawed. Face it, the OSI reference model is over a decade old, and even then it was inadequate to describe real working networks. (As opposed to, say, the paper networks ISO is *still* building...) What's far more important is the relationship between the various protocol modules (i.e., who runs "on top of" who), not which arbitrary "layers" they "belong" to. See Postel and Cohen's paper "The ISO Reference Model and Other Protocol Architectures" for an iron-clad proof that N = N+1 for 1 <= N <= 7, according to the ISO model. My math professors used to tell me that when you get a result like this and you haven't made an error in your logic, then your fundamental assumptions must be wrong... Phil
Stevens@A.ISI.EDU (Jim Stevens) (12/20/87)
Phil, I tried to address 2 problems in my note about protocol reference models: (1) where do things like network management protocols belong in general and (2) identification of a common problem in charts showing the packet radio network protocol layering. It is true that the answer to the first problem depends upon a persons acceptance of the different protocol reference models. However the second problem is independent of whether there are always 7 layers in a network or whether there are N layers in a network or whether layers can be skipped (i.e. null layers). In fact, I have seen this second problem in many different papers using both the ISO/OSI reference model and any of the several slightly different DoD reference models. The second problem is caused when people confuse the host/terminal interface to a network as being the protocols which run within the network. This has especially been true for radio networks (such as Packet Radio and Packet Satellite). I can create an alternate example of this confusion for the Ethernet. My PC is connected to a server on the Ethernet via an RS232 link. If I were to confuse my interface to the Ethernet as being the actual protocols used within the Ethernet, then I would say that the physical protocol used in the Ethernet is RS232 instead of IEEE 802.3. However, most of us would agree that this is incorrect. The reason why the confusion with radio networks continues is because there are many more people who use other networks such as Ethernet than Packet Radio. Thus when most people write a paper and create a chart, they do like I do. They (and I) look at earlier papers and assume that the information in the areas that they do not know is correct. Thus the goal of my earlier message with respect to this second problem was to provide the correct information about what Link and Physical Layer protocols are used within the DARPA Packet Radio Networks, and explain that there is confusion between interface protocols and network protocols so that this type of error is repeated less often. Jim P.S. I do not intend to argue about the absolute correctness of the OSI model for all types of actual networks. (I myself can list examples of things like networks that appear to be just a link to another network which then appears to be just a link to yet another higher network.) But the most important thing about the OSI model is that is has become ubiquitous and is a useful way to describe concepts. I would classify all of the existing network management protocols as being of 2 types: (a) lying within the network layer (ex. ICMP) or (b) lying above the network layer (ex. EGP). This is the same type of structure breakdown that was decided upon in the protocol model work that I referenced in my earlier message. Thus the goal of my earlier message with respect to the this first problem was to indicate that while I agree that EGP, GGP, and HMP lie above the Internet Protocol, I do not consider them to be transport protocols, and provide reference to the good work that is being performed on network management architectures. In particular, I feel that the CCITT I.320 Recommendation on the ISDN Protocol Reference Model contains many useful ideas. -------
chris@GYRE.UMD.EDU (Chris Torek) (12/20/87)
Sure, the ISORM is useful. But like the formula for adding velocities, v_sum = v_1 + v_2 one should remember that it is, in a fundamental sense, wrong, or perhaps better to say *limited*. I imagine what Phil dislikes is that ISORM proponents never mention the limitations. Chris