On Sun, May 30, 2010 at 11:45 PM, Kurt Buff <kurt.b...@gmail.com> wrote: > Well then, to expand upon my thoughts:
Ah, thank you. > Using a model that bears little relationship to reality is a faux pas, > and likely to lead you to bad conclusions. I suppose that's possible. However, I don't think that even OSIs detractors can honestly say it bears little relationship with reality. Indeed, I would argue the concepts it describes are almost always going to be encountered in *any* communications system design, even if only to say it doesn't apply or isn't being implemented. > There are also dangers > involved in adding layers to a conceptual model of networking, as > described in RFC3439 (https://tools.ietf.org/html/rfc3439). That it's specifically talking about implementations, not conceptual models. To quote, "Such ordering constraints are in conflict with efficient *implementation of data manipulation functions*." (Section 3, page 7; emphasis added). Note also that that same criticism is also leveled at the design of IP. "One could accuse the layered model (e.g., *TCP/IP* and ISO OSI) of causing this conflict." (ibid; emphasis added) It then goes on to detail how allowing the IP network layer to be optimized for various datalink transports has been a performance win. I don't see how that supports the idea that "OSI is evil". > It ain't scientific. Claim without a supporting argument. > I believe it's better to acknowledge that everything above layer 3 is > a bunch of different protocols, some of which stand alone and some of > which are encapsulated in other protocols, than to use a flawed model > and fool ourselves into thinking something that isn't so. I don't see how using the OSI model as a common frame of reference thereby means we're going to fool ourselves into thinking something isn't so. (I am ignoring the tautology in your supporting argument of assuming "flawed model".) You mention in another message that OSI has never really been implemented. If that's your concern, I think you are missing a subtle but critical point in my argument (and what I believe are the arguments of others): OSI is useful as an *abstract model*. I don't think anyone here is advocating for a strict implementation of OSI as an actual network stack. However, when we talk about things that do networking, it's extremely advantageous to have common, concise terminology which is shared among all involved. That we can say things like "TCP handles transport and session layer duties; anything higher-level is implementation-specific" and everyone will know what we mean. It's common terminology, not a technological mandate. Even when a given OSI layer is not implemented in a system, that very fact is a key attribute of a design. For example, say you have something designed around Ethernet. If you want to move that to FDDI, you know you just have to adapt your handling of the data link layer. But if you want to move that to EIA-485, you're going to have to implement your own data link layer, because EIA-485 is a physical layer only. Without the common terminology of the OSI model, that discussion would be much more confused. So it's good that OSI specifies so many layers and abstractions, even when things are often not implemented that way -- it gives us clear, concise, common terminology for any number of different scenarios. Adopting the IP model for such conversations would be especially cumbersome, because in the IP model, anything below the network layer is implementation-specific, and anything above the IP protocol layer is implementation specific. Great if all you ever do is work with IP routers, but really limiting if you're building an Ethernet switch or a streaming video protocol. -- Ben ~ Finally, powerful endpoint security that ISN'T a resource hog! ~ ~ <http://www.sunbeltsoftware.com/Business/VIPRE-Enterprise/> ~
