Does ARP Belong to Layer 2 Or Layer 3 OSI Reference Model??? - TCP-IP

I have posted this topic at somewhere else,but nobody could gives me a
definitive answer,so I post it here and hope someone make it clear for
me

iLRainyday writes:

> I have posted this topic at somewhere else,but nobody could gives me a
> definitive answer,so I post it here and hope someone make it clear for
> me

Short answer - it's both.

TCP/IP has NOTHING to do with the OSI reference model.

There's a reason why we use TCP/UDP/IP stack instead of OSI stack.
The OSI reference model is fatally flawed.
It's a classic example on how to design a protocol stack the wrong way.

ARP crosses the layers in order to function. That's what real protocols do.

As Padlipski says, "If you know what you are doing, three layers is
enough; if you don't, even seventeen levels won't help"



--
Posted via a free Usenet account from http://www.teranews.com

On Feb 4, 1:11*pm, Bruce Barnett
wrote:
> iLRainyday writes:
> > I have posted this topic at somewhere else,but nobody could gives me a
> > definitive answer,so I post it here and hope someone make it clear for
> > me
>
> Short answer - it's both.
>
> TCP/IP has NOTHING to do with the OSI reference model.

The original poster was not asking about the OSI stack, but only about
the OSI 7-layer model.

ARP resolves the L2 address from an L3 address, but ARP requests and
replies are not routed. And the L3 address can be for any Layer 3
protocol. ARP is not limited to responding only to IP devices.

I don't think that anyone doubts the applicability of the ISO/OSI 7-
layer model, AT LEAST with respect to layers 1-4. Maybe the usefulness
of layers above 4 might be debated. The question in the subject line
is that ARP operates at Layer 2, in the sense that it does not use IP,
or other L3 headers, for routing purposes. It is broadcast at the L2
layer, and replies are sent back using L2 the address of the
requesting host.

Bert

Albert Manfredi wrote:
> On Feb 4, 1:11?pm, Bruce Barnett
> wrote:
> > iLRainyday writes:
> > > I have posted this topic at somewhere else,but nobody could gives me a
> > > definitive answer,so I post it here and hope someone make it clear for
> > > me
> >
> > Short answer - it's both.
> >
> > TCP/IP has NOTHING to do with the OSI reference model.

> The original poster was not asking about the OSI stack, but only about
> the OSI 7-layer model.

> ARP resolves the L2 address from an L3 address, but ARP requests and
> replies are not routed. And the L3 address can be for any Layer 3
> protocol. ARP is not limited to responding only to IP devices.

> I don't think that anyone doubts the applicability of the ISO/OSI 7-
> layer model, AT LEAST with respect to layers 1-4. Maybe the usefulness
> of layers above 4 might be debated. The question in the subject line
> is that ARP operates at Layer 2, in the sense that it does not use IP,
> or other L3 headers, for routing purposes. It is broadcast at the L2
> layer, and replies are sent back using L2 the address of the
> requesting host.

You have left-out layers 8 and 9:

https://secure.isc.org/index.pl?/store/t-shirt/



rick jones
--
No need to believe in either side, or any side. There is no cause.
There's only yourself. The belief is in your own precision. - Jobert
these opinions are mine, all mine; HP might not want them anyway...
feel free to post, OR email to rick.jones2 in hp.com but NOT BOTH...

On Feb 4, 12:11*pm, Bruce Barnett
wrote:
> iLRainyday writes:
> > I have posted this topic at somewhere else,but nobody could gives me a
> > definitive answer,so I post it here and hope someone make it clear for
> > me
>
> Short answer - it's both.
>
> TCP/IP has NOTHING to do with the OSI reference model.

Hmmm...I guess.

Surely you will agree though, that one can see a bit of layering
regarding what is located at each end of an abstract unit of data
transmitted and received:

1. interface to interface [link-layer frame]
2. node to node [network-layer packet]
3. port to port..[transport layer whatever]
4.+..[There is some irregularity at this point and higher that
researchers still have not figured out]

> There's a reason why we use TCP/UDP/IP stack instead of OSI stack.
> The OSI reference model is fatally flawed.
> It's a classic example on how to design a protocol stack the wrong way.

I definitely agree here. When I read the OSI "specification", it was
like being locked in a closet that smelled like 500-year-old moth
balls. If someone asked me to find the OSI specification within 5
minutes using Google to save my life, I'd probably die.

> ARP crosses the layers in order to function. That's what real protocols do..

Yup. Primitives are still king. You do need those. But OSI put the
cart before the horse. They took something that was good - the notion
of layering - and presumed it would save the world. It might have been
better to write an entire, working protocol stack, think and rethink
it until one is confident that it is nearing its regular form, clean
it up, then write about that. Instead, they created 500-page
documents filled with nauseating vaguery that would make even the most
patient technophile get eye-rot.

> As Padlipski says, "If you know what you are doing, three layers is
> enough; if you don't, even seventeen levels won't help"

Hah! This is classic.

-Le Chaud Lapin-

Albert Manfredi writes:

> ARP resolves the L2 address from an L3 address, but ARP requests and
> replies are not routed. And the L3 address can be for any Layer 3
> protocol. ARP is not limited to responding only to IP devices.

Well, while ARP routes at layer 2, it needs information from layer 3.
That, to my mind, violates the OSI concept of layering.

> I don't think that anyone doubts the applicability of the ISO/OSI 7-
> layer model, AT LEAST with respect to layers 1-4.

Applicability? It's a conceptual model that lawnchair theorists like to use.
Those that have tried to build protocols based on the concept
failed, as it caused increased complexity and decreased efficiency.

If the concept of layering is so wonderful, where are these layered
protocols used?




--
Posted via a free Usenet account from http://www.teranews.com

On Feb 5, 6:23*am, Bruce Barnett
wrote:

> Well, while ARP routes at layer 2, it needs information from layer 3.
> That, to my mind, violates the OSI concept of layering.

The layer 3 information in ARP frames, i.e. the L3 address, is nothing
but data payload in ARP. So it really doesn't violate anything.

> > I don't think that anyone doubts the applicability of the ISO/OSI 7-
> > layer model, AT LEAST with respect to layers 1-4.
>
> Applicability? It's a conceptual model that lawnchair theorists like to use.
> Those that have tried to build protocols based on the concept
> failed, as it caused increased complexity and decreased efficiency.

I've always found these arguments hard to parse.

I've used this example before: If someone wants to learn how
automobile drivelines work, he will probably be taught a generic
model, which includes engine, transmission, differential, wheels.

Now, if some car designs combine the transmission and differential in
one unit called "transaxle," does this invalidate the model? Of course
not.

Same applies here. The 7-layer model doesn't HAVE to be treated as a
straightjacket. But at the same time, someone who doesn't understand
the difference between L1, L2, L3, and L4 (at the very least) is
someone who is completely lost about the topic of digital networks.

Bert

On Feb 5, 3:23 am, Bruce Barnett
wrote:

> Well, while ARP routes at layer 2, it needs information from layer 3.
> That, to my mind, violates the OSI concept of layering.

Nonsense. ARP is built on top of layer 2, as a layer 3 protocol should
be. ARP is a layer 3 protocol. ARP implements layer 3 services by
using a layer 2 protocol. That is basically how layering is supposed
to work. A layer N service provides layer N capabilities by using a
layer N-1 service.

> If the concept of layering is so wonderful, where are these layered
> protocols used?

Everywhere. Ethernet is a layer 2 protocol, and it is often helpful to
understand it by thinking of it in that way.

Just don't eat the recipe.

DS

On Feb 5, 7:06*pm, David Schwartz wrote:

> Nonsense. ARP is built on top of layer 2, as a layer 3 protocol should
> be. ARP is a layer 3 protocol. ARP implements layer 3 services by
> using a layer 2 protocol. That is basically how layering is supposed
> to work. A layer N service provides layer N capabilities by using a
> layer N-1 service.
>
> > If the concept of layering is so wonderful, where are these layered
> > protocols used?
>
> Everywhere. Ethernet is a layer 2 protocol, and it is often helpful to
> understand it by thinking of it in that way.
>
> Just don't eat the recipe.

I agree with all except the first part.

IP implements L3 services by being layered over L2 protocols. By doing
so, IP routes packets between L2 networks, which is the definition of
a L3 service.

But ARP does no such thing. It only carries the IP address within a L2
network. In a somewhat similar way, IGMP and ICMP operate directly
over IP, but they don't provide L4 services. They only enable a L3
function to take place, e.g. multicast, and/or they provide L3
troubleshooting tools. They don't provide L4 services in any way
comparable with TCP or UDP.

Anyway, Douglas Comer, while not being very specific, does say that
ARP is a very low level protocol.

Bert

On Feb 5, 8:35*pm, Albert Manfredi wrote:
> On Feb 5, 7:06*pm, David Schwartz wrote:
>
> > Nonsense. ARP is built on top of layer 2, as a layer 3 protocol should
> > be.
>
> I agree with all except the first part.

Perhaps the disconnect here is that I am responding to the question
"at what layer does ARP operate?" Maybe "belong to" is more ambiguous,
since ARP does carry L3 information as part of its payload.

IP operates at L3. ARP operates at L2.

Bert

Split the difference - call ARP a Layer 2.5 protocol.

rick jones
--
No need to believe in either side, or any side. There is no cause.
There's only yourself. The belief is in your own precision. - Jobert
these opinions are mine, all mine; HP might not want them anyway...
feel free to post, OR email to rick.jones2 in hp.com but NOT BOTH...

On Feb 5, 8:02*pm, Rick Jones wrote:
> Split the difference - call ARP a Layer 2.5 protocol. *

Funny. That's the image I get when I think about ARP.

I see both points being made, and I would have to slightly lean toward
Albert's point of view, but OTOH, David is right about something I
think is slightly different than what we are talking about - which is
upper layer protocols employ lower layer protocols to achieve certain
effects, but there is also a bootstrapping that goes on in protocol
design that makes it impossible to regard a protocol stack as strictly
conforming to layers.

It would be nice if someone were to write a paper describing this
bootstrapping phenomenon because it is prevalent in all kinds of
systems, not just software.

For example:

1. If a mailman delivers mail to an area of a city that includes his
own, does that means he delivers mail to himself? Does he then put it
in his own mailbox and take it right back out as required by law or
does he put it in his pocket?

2. If a neurosurgeon feels tingling sensation in his frontal lobe that
warns of impending danger, does he prescribe preemptive medication or
does he see a doctor?

3. The memory manager of an OS allocates memory to processes that need
it. But the memory manager needs memory of its own to function. From
where should it request memory?

4. A sheriff in a northern USA state recently wrote himself a ticket
for his role in minor traffic accident He imposed the appropriate
fine, plead no-contest, and paid the fine, even though it was probably
case where two individuals involved might have just settled it between
themselves (http://tinyurl.com/3xxm99). .

All of these examples demonstrate lateral-self-service principle of
systems, whereby one layer of abstraction of a system will not only
provide services to conceptually higher layers, but will also provide
services to lateral portions of the system, including itself.

Perhaps this is one of the points that David was making. If so, I
agree. I think that the secret to finding a well-formed protocol
stack (beyond TCP/IP) lies in acknowledging and embracing this notion,
but not getting so carried away with it that one makes a mess.
Minimalism is key, so that if you adhere to this philosophy with a
minimalist approach, the areas of necessary applicability will become
obvious and singular in their manifestation.

-Le Chaud Lapin-

On Feb 5, 5:35 pm, Albert Manfredi wrote:

> IP implements L3 services by being layered over L2 protocols. By doing
> so, IP routes packets between L2 networks, which is the definition of
> a L3 service.

Right.

> But ARP does no such thing. It only carries the IP address within a L2
> network.

But this:

1) Implemented using L2 services, and

2) Needed to implement the IP L3 services.

This makes it part of L3.

> In a somewhat similar way, IGMP and ICMP operate directly
> over IP, but they don't provide L4 services.

I agree. They provide L3 services. They are part of the IP L3 suite.

> They only enable a L3
> function to take place, e.g. multicast, and/or they provide L3
> troubleshooting tools. They don't provide L4 services in any way
> comparable with TCP or UDP.

I agree. IGMP and ICMP live entirely at L3.

> Anyway, Douglas Comer, while not being very specific, does say that
> ARP is a very low level protocol.

It's a part of the IP L3 protocol.

DS

Hello,

Albert Manfredi a écrit :
>>On Feb 5, 7:06 pm, David Schwartz wrote:
>>
>>>Nonsense. ARP is built on top of layer 2, as a layer 3 protocol should
>>>be.

Not everything that is built on top of layer N is layer N+1 (and
conversely). IIRC the protocol layer is defined by the services this
protocol offers, not by which protocol layer it runs on.

> Perhaps the disconnect here is that I am responding to the question
> "at what layer does ARP operate?" Maybe "belong to" is more ambiguous,
> since ARP does carry L3 information as part of its payload.
>
> IP operates at L3. ARP operates at L2.

What about IPv6 neighbor discovery ? It does the same job as ARP and
runs on top of ICMPv6.

On Feb 6, 1:27 am, Pascal Hambourg
wrote:

> Not everything that is built on top of layer N is layer N+1 (and
> conversely). IIRC the protocol layer is defined by the services this
> protocol offers, not by which protocol layer it runs on.

The service ARP provides is part of the L3 service. Don't think of ARP
as a protocol by itself, think of it as part of the IP service.

DS

"David Schwartz" wrote in message
news:39b660c7-4ef3-4cc8-9a63-57a05027f091@v4g2000hsf.googlegroups.com...
> On Feb 6, 1:27 am, Pascal Hambourg
> wrote:
>
>> Not everything that is built on top of layer N is layer N+1 (and
>> conversely). IIRC the protocol layer is defined by the services this
>> protocol offers, not by which protocol layer it runs on.
>
> The service ARP provides is part of the L3 service. Don't think of ARP
> as a protocol by itself, think of it as part of the IP service.
>
> DS

ARP is a layer 2 protocol
http://tools.ietf.org/html/rfc826
http://en.wikipedia.org/wiki/Address...ution_Protocol

iLRainyday writes:
> I have posted this topic at somewhere else,but nobody could gives me a
> definitive answer,so I post it here and hope someone make it clear for
> me

osi layer 2 is point-to-point inferface

osi layer 3 is network layer ... figuring out which point-to-point
interface to use

LANs, MACs, ARP, etc ... aren't covered by OSI reference model ... the
physical interface has some of the characteristics of a point-to-point
interface. however, the networking characteristics of LANs can be
considered to place them in the middle of networking stuff defined in
OSI layer 3.

ISO compounded the problems with the OSI reference model by passing out
an edict that ISO networking standards (and work by ISO nationally
chartered bodies) could only be done on protocols that conformed to the
OSI reference model (which was also a ISO standard).

I was involved in trying to get HSP (high-speed protocol) work done in
X3S3.3 (the US ISO chartered standards body responsible for standards
work related to OSI layers 3&4).

eventually HSP standards work couldn't be considered in x3s3.3 (and ISO)
because it violated OSI reference model:

1) went directly from transport interface to the mac interface bypassing
network interface ... violating OSI reference model

2) supported "internetworking" layer (i.e. IP) ... something that
doesn't exist in the OSI reference model ... and violates the OSI
reference model (aka may require sending traffice to a "gateway"
responsible for internetworking between networks ... not having
internetworking ... there are also no gateways).

3) interfaced to LAN/MAC ... which doesn't exist in OSI reference model,
and therefor violates the OSI reference model.

ARP is part of the LAN/MAC genre not defined in OSI ... so it takes a
lot of really hard squinting to pretend to see it fitting.

misc. past posts
http://www.garlic.com/~lynn/subnetwork.html#xtphsp

eager wrote:

> ARP is a layer 2 protocol

ARP translates layer 3 addresses into layer 2 addresses. As such, it
is part of the implementation of layer 3 using layer 2. I don't see
any of those links claiming ARP is a layer 2 protocol, in fact, they
claim the reverse.

The layer 2 protocol will work just fine without ARP.

DS

"eager" wrote:
> ARP is a layer 2 protocol
> http://tools.ietf.org/html/rfc826
> http://en.wikipedia.org/wiki/Address...ution_Protocol

I disagree:

Does Ethernet need ARP in order for it to deliver Ethernet frames?
No.

Does IP need ARP in order for it to deliver IP datagrams over Ethernet?
Yes.

The OSI Reference Model is just a specific way of doing classic functional
decomposition of all the tasks needed for multiple computers to communicate
in real world situations. The model adopted a sequential or layering
approach. One of the reasons the OSI "reference model" was created was to
aid in communication among humans working on networking software - after
all, computers currently have no use for an abstract "model" or have any
need to "reference" back to that model during their sending and receiving
of packets.

I would suggest that one way to determine the lowest layer at which a
particular function or task is best categorized is to determine the lowest
layer at which the following question can be answered yes:

Does this OSI layer need X in order for it to accomplish any of the tasks
required of it in the OSI Reference Model?

If the reference model is sufficiently complete, I believe one should be
able to fit all of TCP/IP functionality into the OSI Reference Model using
that sort of mapping strategy. If the model is complete enough, the reverse
mapping to an actual implementation like TCP/IP may not always be possible
or relevant. The forward mapping is sometimes useful from both an
educational standpoint and from an engineering standpoint when one needs to
compare communications "stacks".

All the above is in my humble opinion, of course.

David Schwartz wrote:
> eager wrote:

> > ARP is a layer 2 protocol

> ARP translates layer 3 addresses into layer 2 addresses. As such, it
> is part of the implementation of layer 3 using layer 2. I don't see
> any of those links claiming ARP is a layer 2 protocol, in fact, they
> claim the reverse.

> The layer 2 protocol will work just fine without ARP.

Layer 3 protocols can also work just fine without ARP. It might not
be as convient, but it can work. ARP also isn't the only thing out
there ever to translate IP to MAC addresses. "Back in the day" HP had
something called Probe for its NS Transport which was used to resolve
names to IPs, IPs to MACs and also exchange routing information..

I suppose if folks cannot agree on whether it is a floor wax or a
dessert topping they could just agree to call it Shimmer...

rick jones
--
portable adj, code that compiles under more than one compiler
these opinions are mine, all mine; HP might not want them anyway...
feel free to post, OR email to rick.jones2 in hp.com but NOT BOTH...