Hey! Keep Your Hands Out Of My Abstraction Layer! - TCP-IP

This is a discussion on Hey! Keep Your Hands Out Of My Abstraction Layer! - TCP-IP ; Rich Seifert wrote: > In article , > "Le Chaud Lapin" wrote> > > I have never seen an OSI stack. I do not doubt that some vendors put > > something in a box and shipped it to customers. ...

+ Reply to Thread
Page 6 of 6 FirstFirst ... 4 5 6
Results 101 to 108 of 108

Thread: Hey! Keep Your Hands Out Of My Abstraction Layer!

  1. Re: Hey! Keep Your Hands Out Of My Abstraction Layer!


    Rich Seifert wrote:
    > In article <1148458415.349242.150380@u72g2000cwu.googlegroups. com>,
    > "Le Chaud Lapin" wrote>
    > > I have never seen an OSI stack. I do not doubt that some vendors put
    > > something in a box and shipped it to customers. What I doubt is that
    > > it was actually a protocol stack.
    > > For example, what was the format of the network-layer addresses? What
    > > was the API for the programmer? What was in between? These are
    > > fundamental questions that are *not* answered by OSI. And if they are
    > > not answered by OSI, how can one claim OSI compatibility?
    > >

    >
    > There are two parts to what most people refer to as "OSI". First is the
    > layered *model*. That is the "abstraction" that I believe you are
    > referring to. That model is defined in ISO Standard 7498. It is not some
    > "vague, warm-and-fuzzy concept", but a formalized, standard model for
    > network protocol operation. I recommend you get a copy of the standard
    > and read it; it may be eye-opening to see "the real McCoy". The ISO 7498
    > (OSI Model) standard does not define any specific protocols to fit into
    > the layered model it presents.
    >
    > However, a number of other ISO standards *DO* specify such protocols,
    > i.e., message formats, semantics, and rules for behavior of conformant
    > implementations. This is the second part of "OSI".


    [snippage]

    Ok. I downloaded the OSI specification.

    There is one recurring theme that is present throughout the
    "specification". I quote:

    4.3.5 "It should be emphasized that the Basic Reference Model does not,
    by itself, specify the dtailed and precise functioning of the open
    system and, therefore, it does not specify the external behavior of
    real open systems and does not imply the structure of the
    implementation of a real open system."

    Again...

    4.36 "The reader not familiar with the technique of abstract modelling
    is cautioned that those concepts introduced in the description of open
    systems constitute and abstraction despite a similar appearance to
    concepts commonly found in real systems. Therefore, real open systems
    need not be implemented as described by the model."

    I smell an abstraction whore.

    After reading through this document, which, I must point out, had a
    sedative effect more powerful than any drug I've ever taken, I
    suspicions were only confirmed even further. OSI, the model, is too
    vague. It is a generic description of how to do networking under a
    layered-model. OSI "stacks", the implementations, fall into one of two
    categories:

    1. That which is new, which was inferior to other "stacks".
    2. That which is old, which OSI attempted to pimp as its own (TCP/IP
    for example).

    This is intellectual dishonesty. Documents like OSI/IEC 7498-1 hurt
    more than help. To the uninitiated, it gives the impression that there
    is "meat" in what is written. After all, if you are a venture
    capitalist or government grants administrator with your finger on
    $50,000,00US, and you know relatively little about computer
    networking, 68 pages of fluffy, vague, amorphous babble looks pretty
    good. And if you can appropriate something that already exists (X.25,
    DECNET, TCP/IP), it looks even better. So you sign the
    disbursement/grant and wait. Meanwhile, those that are "not OSI
    compliant" have two good reasons to "become OSI compliant".

    1. Everyone else is doing it, better not get left out.
    2. New purchases from old customers are guaranteed (US Government for
    eg.)

    So even though people who make DECNET/X.25/etc. know that it's a crock
    of poo, they all join in on the bandwagon, because there are only
    winners if they do, no losers. The people who know that TCP/IP is
    "better" and become angry, naturally, for fear that truth will be lost
    in the confusion. No one likes 100 megabits of caca, after all.

    Bottom line: anyone can write a vague description of how things should
    be, then proclaim that "our vague description will be embodied
    shortly." I could probably write up a vague description of how an
    atomic bomb works, or how to get to Mars.

    The devil is in the details.

    TCP/IP "won" because it was more real than OSI ever was.

    -Le Chaud Lapin-


  2. Re: Hey! Keep Your Hands Out Of My Abstraction Layer!

    In article ,
    Jeff Liebermann wrote:

    >>I've heard similar blather for a new reinvention of TCP sometime in the
    >>last 6 or 12 months, again supposedly for fixing the problems in TCP
    >>that supposedly make TCP completely unusable over radios. However, I
    >>can't recall the name for the new effort.

    >
    >Not TCP/IP but the MAC layer for mesh networks.
    > http://portal.acm.org/citation.cfm?id=1080847
    >There are actually a few more such proposals but I can't find them
    >right now.


    No, those are not what I meant. Link layer (what is a "MAC layer"?--I'm
    being picking to make a point) ad hoc meshes differ from TCP/IP. Strange
    new link layers are not utterly ignorant replacements of transport
    protocols like WAP. Never mind what the bogus experts might be saying
    about the latest link layer replacing TCP. They've said that about
    every new link layer standardized since 1985. They haven't any real
    clues except the 7 layer posters what might make a layer. That they
    say such nonsense doesn't keep everyone else from thinking about running
    IP (or CLNP, or even CONS) over the new link layer, and TCP (or TP1
    through TP4) above IP (or one of the ISO OSI protocols).


    Vernon Schryver vjs@rhyolite.com

  3. Re: Hey! Keep Your Hands Out Of My Abstraction Layer!

    "Vernon Schryver" wrote:

    > Yes, but there are many self-described network types who clearly can't
    > differenticate between link and transport layers except by rote,
    > uncomprehending reference to examples in the layer posters and who
    > proceed to dictate how to do this or that. For example, today someone
    > wrote in an unrelated newsgroup:
    >
    > } As a protocol SMTP is a peer to peer protocol; thus ISPs have to
    > either
    > } block port 25 or resort to Layer 7 firewalling which would block
    > SMTP
    > } regardless of which port is used.
    >
    > Never mind whether SMTP is properly a peer to peer protocol. Also
    > never
    > mind the definition of "email" as distinguished from other
    > applications.
    > Is x.400 "email"?--I suspect not by that author's unconscious
    > definition.
    > Ignore how little SMTP involves one of the official SMTP well konwn
    > ports. Just notice the incantation of "Layer 7" as if it lets a
    > firewall
    > distinguish "email" from other text applications without also using
    > the
    > fact that a well known SMTP port is somehow involved.


    Actually, I thought the point was rather obvious. The so-called
    "incantation" was pointing out that you had to go higher in the protocol
    stacks (dig deeper into the packet) than what a network normally does,
    to protect the network. Just as IGMP snooping is a link layer technique
    that looks higher in the protocol stack (i.e. deeper into a packet) than
    the link layer normally does, to accomplish a link layer function. (As
    opposed to GMRP, which is therefore useful for more than just IP
    Multicast over that link layer.)

    But this is digressing.

    Essentially, I turn your comment around. I agree that professional
    meeting attenders and clueless marketers babble on about layers without
    understanding what they're talking about. Witness the drivel about layer
    x switches. But my suggestion to you is, anyone who doesn't understand
    about protocol layering, anyone who insists on confusing the OSI
    protocol layer model with the ISO/OSI protocols themselves, is either a
    rank amateur or a charlatan, or possibly both.

    Once again, there is no mention of ISO or OSI in RFC 791, and yet the
    layered protocol model is clearly used. RFC 791 is the current standard
    for *IP*, not ISO/OSI protocols. It may only show 4 layers vs 7, but
    anyone with the slightest clue can see it accomplishes exactly the same
    function, and is totally consistent with the OSI 7-layer model.

    Bert


  4. Re: Hey! Keep Your Hands Out Of My Abstraction Layer!


    Albert Manfredi wrote:
    > Once again, there is no mention of ISO or OSI in RFC 791, and yet the
    > layered protocol model is clearly used. RFC 791 is the current standard
    > for *IP*, not ISO/OSI protocols. It may only show 4 layers vs 7, but
    > anyone with the slightest clue can see it accomplishes exactly the same
    > function, and is totally consistent with the OSI 7-layer model.


    while layering is consistent with other architectures that are layered
    .... it contains a "internetworking" layer that is totally absent from
    the OSI 7-layer model.

    this is one of the things that we ran afoul of in ansi/iso in the late
    80s and early 90s with high-speed protocol. ISO (and iso-chartered
    national standards body) would not standardized protocols that were not
    in comformance with the OSI 7-layer model. Since the OSI 7-layer model
    doesn't have an "internetworking" layer (an abstractions that sits
    somewhere between the bottom of transport and the top of networking ...
    where the OSI 7-layer model shows transport going directly to
    networking and having absolutely no provisions for internetworking).
    http://www.garlic.com/~lynn/subnetwork.html#xtphsp

    i've had similar discussions about the internal network. the internal
    networking nodes had a form of gateway layer built into the majority of
    the nodes.
    http://www.garlic.com/~lynn/subnetwork.html#internalnet

    while arpanet had much more traditional appearing network (i.e. much
    more coformance with osi) implementation in the 70s. it didn't really
    see the gateway layer until the change-over to internetworking on
    1/1/83.
    http://www.garlic.com/~lynn/subnetwork.html#internet

    my assertion is that one of the reasons that the internal network was
    so much larger than the arpanet from just about the beginning until
    sometime mid-85 .... was that the internal network had something like a
    gateway layer from the beginning which the arpanet didn't get until the
    change-over to internetworking on 1/1/83.

    arpanet had possible 250 hosts on 1/1/83 ... but it may have actually
    been as little as 100 hosts. there were on the order of 100 arpanet
    "IMP" nodes on 1/1/83 .... and all sub-addresses appeared to have been
    referred to as "hosts". there may have been as many as 100 of the 250
    "hosts" (on 1/1/83) might have possible been terminal controllers
    rather than more generalized dataprocessing machines (i.e. computers).
    refs:
    http://www.garlic.com/~lynn/2006k.html#40 Arpa address
    http://www.garlic.com/~lynn/2006k.html#42 Arpa address

    by comparison, the internal networking had nearly 1000 host/nodes on
    1/1/83
    (it officially passed 1000 nodes in jun83)
    http://www.garlic.com/~lynn/2006k.html#43 Arpa address
    http://www.garlic.com/~lynn/2006k.html#8 Arpa address

    there was a secondary issue with the internet over taking the internal
    network in number of nodes. there was a strong corporate push to
    restrict PCs and workstations to terminal emulation on the internal
    network .... where they were becoming full-fledge networking nodes on
    the internet. recent posts on the subject
    http://www.garlic.com/~lynn/2006k.html#9 Arpa address
    http://www.garlic.com/~lynn/2006k.html#21 Sending CONSOLE/SYSLOG To
    Off-Mainframe Server
    http://www.garlic.com/~lynn/2006k.html#25 Can anythink kill x86-64?


  5. Re: Hey! Keep Your Hands Out Of My Abstraction Layer!


    Albert Manfredi wrote:
    > Once again, there is no mention of ISO or OSI in RFC 791, and yet the
    > layered protocol model is clearly used. RFC 791 is the current standard
    > for *IP*, not ISO/OSI protocols. It may only show 4 layers vs 7, but
    > anyone with the slightest clue can see it accomplishes exactly the same
    > function, and is totally consistent with the OSI 7-layer model.


    while layering is consistent with other architectures that are layered
    .... it contains a "internetworking" layer that is totally absent from
    the OSI 7-layer model.

    this is one of the things that we ran afoul of in ansi/iso in the late
    80s and early 90s with high-speed protocol. ISO (and iso-chartered
    national standards body) would not standardized protocols that were not
    in comformance with the OSI 7-layer model. Since the OSI 7-layer model
    doesn't have an "internetworking" layer (an abstractions that sits
    somewhere between the bottom of transport and the top of networking ...
    where the OSI 7-layer model shows transport going directly to
    networking and having absolutely no provisions for internetworking).
    http://www.garlic.com/~lynn/subnetwork.html#xtphsp

    i've had similar discussions about the internal network. the internal
    networking nodes had a form of gateway layer built into the majority of
    the nodes.
    http://www.garlic.com/~lynn/subnetwork.html#internalnet

    while arpanet had much more traditional appearing network (i.e. much
    more coformance with osi) implementation in the 70s. it didn't really
    see the gateway layer until the change-over to internetworking on
    1/1/83.
    http://www.garlic.com/~lynn/subnetwork.html#internet

    my assertion is that one of the reasons that the internal network was
    so much larger than the arpanet from just about the beginning until
    sometime mid-85 .... was that the internal network had something like a
    gateway layer from the beginning which the arpanet didn't get until the
    change-over to internetworking on 1/1/83.

    arpanet had possible 250 hosts on 1/1/83 ... but it may have actually
    been as little as 100 hosts. there were on the order of 100 arpanet
    "IMP" nodes on 1/1/83 .... and all sub-addresses appeared to have been
    referred to as "hosts". there may have been as many as 100 of the 250
    "hosts" (on 1/1/83) might have possible been terminal controllers
    rather than more generalized dataprocessing machines (i.e. computers).
    refs:
    http://www.garlic.com/~lynn/2006k.html#40 Arpa address
    http://www.garlic.com/~lynn/2006k.html#42 Arpa address

    by comparison, the internal networking had nearly 1000 host/nodes on
    1/1/83
    (it officially passed 1000 nodes in jun83)
    http://www.garlic.com/~lynn/2006k.html#43 Arpa address
    http://www.garlic.com/~lynn/2006k.html#8 Arpa address

    there was a secondary issue with the internet over taking the internal
    network in number of nodes. there was a strong corporate push to
    restrict PCs and workstations to terminal emulation on the internal
    network .... where they were becoming full-fledge networking nodes on
    the internet. recent posts on the subject
    http://www.garlic.com/~lynn/2006k.html#9 Arpa address
    http://www.garlic.com/~lynn/2006k.html#21 Sending CONSOLE/SYSLOG To
    Off-Mainframe Server
    http://www.garlic.com/~lynn/2006k.html#25 Can anythink kill x86-64?


  6. Re: Hey! Keep Your Hands Out Of My Abstraction Layer!

    lynn@garlic.com wrote:
    > Albert Manfredi wrote:
    > > Once again, there is no mention of ISO or OSI in RFC 791, and yet the
    > > layered protocol model is clearly used. RFC 791 is the current standard
    > > for *IP*, not ISO/OSI protocols. It may only show 4 layers vs 7, but
    > > anyone with the slightest clue can see it accomplishes exactly the same
    > > function, and is totally consistent with the OSI 7-layer model.

    >
    > while layering is consistent with other architectures that are layered
    > ... it contains a "internetworking" layer that is totally absent from
    > the OSI 7-layer model.


    We disagree.

    The internetworking layer, also shown clearly in Figure 2 of RFC 791,
    is the Network Layer in the OSI model. Layer 3.

    Back in 1980, they called the layer 2 net the "local network," and the
    layer that ties these "local networks" together the "internetworking
    layer." It's very obvious in RFC 791. The top of Section 2.3 says:

    "The function or purpose of Internet Protocol is to move datagrams
    through an interconnected set of networks. This is done by passing the
    datagrams from one internet module to another until the destination is
    reached."

    I don't see how this could be clearer, especially in view of the fact
    that the "local networks" RFC 791 talks about were quite often
    Ethernets, already by 1981. The same type of broadcast LANs one takes
    for granted in OSI Layer 2.

    I think most of the confusions you bring up were caused by meeting
    attenders getting too wrapped up in the specific nature of the "local
    network" technologies they saw, perhaps, the same way as you might get
    confused when IP is layered over ATM or Frame Relay. To me, that's only
    a problem of perspective. I agree that there are, or could be, "layer
    3-like" functions going on in ATM nets. E.g. NHRP (RFC 2332).

    But the sign of grasping a subject, any subject, is to be able to get
    past these conceptual hurdles without losing the picture.

    Bert


  7. Re: Hey! Keep Your Hands Out Of My Abstraction Layer!

    Vernon Schryver wrote:

    (snip)

    > Recall that I wrote a few messages ago that people who actually
    > understand the concept of "network layer" find fewer occasions to use
    > the word "layer" than those who don't. Then notice that the string
    > "layer" does not occur in RFC 791 (although it does occur in RFC 793).
    > Recall also that "The Basic Reference Model for Open Systems
    > Interconnection" was not published until 1983, two years after the
    > 1981 date of RFC 791 and RFC 793. Recall the determinations that the
    > DDN Protocol Suite was unacceptable in part because it did not sufficiently
    > fit the 7 Layer Model.


    There is a story, I believe from a reliable source though I am not sure
    where, that the seven layers were a compromise between six and eight,
    from the US and European (I don't remember which was which) sides.

    Also, that the 48bytes for ATM is a compromise between 32 and 64 (why
    not the geometric mean? it would be more fair.), again between US and
    Europe.

    -- glen


  8. Re: Hey! Keep Your Hands Out Of My Abstraction Layer!

    Vernon Schryver wrote:

    > Recall that I wrote a few messages ago that people who actually
    > understand the concept of "network layer" find fewer occasions to use
    > the word "layer" than those who don't. Then notice that the string
    > "layer" does not occur in RFC 791 (although it does occur in RFC 793).


    What the heck, I'll belabor for a minute.

    Although you sound like you're disagreeing, the disagreements are minor
    and mostly subjective.

    People might not harp on the layer definitions, but in my experience
    that's because those who is actually contribute to the IP evolution
    process have understood the concept and internalized it in all their
    work. It's no big deal, no longer flame bait.

    I just happened to be perusing RFC 4082, written about 1 year ago, when
    I saw your post. So I thought, surely the concept is embedded here as
    it is everywhere else. And it is, first in Section 2:

    "TESLA can be used in the network layer, in the transport layer, or in
    the application layer." These are the OSI names for these layers. You
    don't see any "internetworking layer," right?

    And Section 4 explains the pros and cons of conducting TESLA at each
    layer. Again, read it and see how an understanding of what these layers
    do is taken for granted.

    > Recall also that "The Basic Reference Model for Open Systems
    > Interconnection" was not published until 1983, two years after the
    > 1981 date of RFC 791 and RFC 793.


    Recall that in an earlier reply, I said that the OSI 7-layer model was
    not created out of nothing? And that RFC 760 already proposed a layered
    protocol model in 1980? And I didn't search further back, but wouldn't
    be surprised to find other examples of this. Of course the ISO/OSI
    layers post-date the layers originally assumed in the IP RFCs. IP
    started several years earlier than ISO/OSI. What would one expect?

    > Saying today that TCP/IP fits the Model seems
    > reasonable, but it is revisionism to pretend that the GOSIP battles
    > never happened. For some people like me who suffered through them, it
    > is down right offensive.


    Thing is, I'm not the one still fighting the Civil War. I never
    pretended the flame wars of 10+ years ago never happened. I only assert
    that the OSI 7-layer MODEL has become entrenched in all network
    development, and I say this because I keep seeing people confuse the
    ISO/OSI layered model with the ISO/OSI protocols.

    So, I don't think there's any disagreement here, but I sense that
    others will interpret your posts as disagreements.

    By the bye, just parenthetically, isn't the IS-IS routing protocol,
    which is used in IP nets now, RFCs 1142, 1195 (with IP), 2763, 2966,
    etc. etc., a legacy of the ISO/OSI protocol suite?

    Bert


+ Reply to Thread
Page 6 of 6 FirstFirst ... 4 5 6