[GRLUG] Poll: OSI Model/Protocol

[Adam Tauno Williams]

On Sun, 2011-06-05 at 19:57 -0400, Eric Beversluis wrote:
> I'd like to conduct a poll.
> Having exerted some (not very significant) level of effort, a number of
> years ago, acquainting myself with the OSI model/standard, I'm curious
> if I'm the only one who sees imposing this learning task on computer
> students as pointless.

Define "acquainting"?  I think it is work 'acquainting' with; I don't
know that it merits "study".  It is important to get across the concept
that systems are a stack of abstractions - and that these abstractions
server a virtuous purpose.  It maybe actually *easier* to teach this
with a hypothetical model rather than a 'real-world one'.  The
real-world is so damn tedious.

> Item: in about 8 years since studying all that, I have seen maybe one or
> two references to the model and, as nearly as I can recall, none of the
> work I've done in this time has needed to utilize knowledge of the
> model.

No, but you, even without consciously thinking about it, operate with
the premise that there is a model.

> Item: the de facto standard for networking, TCP/IP, uses a significantly
> different model; nor have I seen any need to understand the TCP/IP
> layers in my work.

I disagree completely and totally.  Being a network or systems
administrator absolutely requires understanding the commonly used model.
We even use it when discussing switches:  there are Layer 1, 2, and 3
switches - that is a real-world application of the model.

Not to sound arrogant... but I think I'm more successful at debugging
and figuring out gnarly network and other IT issues that allot of the
people I've cooperated with over the years.  I attribute this not to
inherent skills but just to being systematic and using the model. 

Also a fair amount of software is terrible, and interoperability
problems occur, because developers *clearly* don't really understand the
technology that are piling their crap on top of.

[Aside: the worst sin in troubleshooting is coming to a 'conclusion' to
soon.  You'll waste time trying to fix the problem to match your
conclusion. Most people who get jammed up on a problem have simply moved
way too fast.  Slow down. Start at the beginning.  Gather information.
And work the model from the bottom up.  Every time.  If you spend any
time with Cisco's, quite good. training documents this is what they tell
you over and over again.  And it works.]

> I would guess designers and builders of hardware and certain programs
> would use knowledge of the TCP/IP layer protocols. But other than these
> users, who out there has had the need to be acquainted with the OSI
> model?

At least administrators and developers [yes, of applications].  What
application doesn't use the network?  They can at least then look at a
wireshark capture and then see that, hey, my application is sending this
exact same request 164 times [true story, BTW].   It's a good final test
that your application isn't stupid.  And selecting SCTP over TCP or how
TCP uses the sliding window really can help you build applications that
perform better.  Or what does HTTP pipelining actually mean?  A web
developer can't really understand that without understanding how TCP
operates.


-- 
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.