Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.os.linux.networking > #3699 > unrolled thread

networking and fragments

Back to article view | Back to comp.os.linux.networking

Contents

  networking and fragments Popping mad <rainbow@colition.gov> - 2016-04-12 00:30 +0000
    Re: networking and fragments William Unruh <unruh@invalid.ca> - 2016-04-12 03:16 +0000
    Re: networking and fragments Pascal Hambourg <boite-a-spam@plouf.fr.eu.org> - 2016-04-23 12:05 +0200
      Re: networking and fragments Jorgen Grahn <grahn+nntp@snipabacken.se> - 2016-04-24 12:53 +0000

#3699 — networking and fragments

Hi

I'm reading this line in the Ross - Kause textbook "Computer Networking - 
Top-Down Aproach" and it says


"The solution is to fragment the data in the IP datagram into two or more 
smaller IP datagrams, encapsulate each of these smaller IP datagrams in a 
separate link-layer frame; and send these frames over the outgoing link. 
Each of these smaller data-grams is referred to as a fragment.

Fragments need to be reassembled before they reach the transport layer at 
the destination. Indeed, both TCP and UDP are expecting to receive 
complete, unfragmented segments from the network layer."

I thought that was wrong.  They have to be reassembled AT the 
destination, not before that.


?

Where am I confused, or is this in fact incorrect?

Reuvain

[toc] | [next] | [standalone]

#3701

On 2016-04-12, Popping mad <rainbow@colition.gov> wrote:
> Hi
>
> I'm reading this line in the Ross - Kause textbook "Computer Networking - 
> Top-Down Aproach" and it says
>
>
> "The solution is to fragment the data in the IP datagram into two or more 
> smaller IP datagrams, encapsulate each of these smaller IP datagrams in a 
> separate link-layer frame; and send these frames over the outgoing link. 
> Each of these smaller data-grams is referred to as a fragment.
>
> Fragments need to be reassembled before they reach the transport layer at 
> the destination. Indeed, both TCP and UDP are expecting to receive 
> complete, unfragmented segments from the network layer."
>
> I thought that was wrong.  They have to be reassembled AT the 
> destination, not before that.

YOu have misread. "need to be reassembled before they reach the
transport layer" is the key phrase. Ie, they are reassembled on the
destination machine but before they are handed over to the transport
layer. 
>
>
> ?
>
> Where am I confused, or is this in fact incorrect?
>
> Reuvain

[toc] | [prev] | [next] | [standalone]

#3713

Popping mad a écrit :
> 
> "Fragments need to be reassembled before they reach the transport layer at 
> the destination. Indeed, both TCP and UDP are expecting to receive 
> complete, unfragmented segments from the network layer."
> 
> I thought that was wrong.

You are right : UDP uses datagrams, not segments.

> They have to be reassembled AT the destination, not before that.
> 
> Where am I confused, or is this in fact incorrect?

IPv4 does not preclude reassembly before the destination. IPv6 does (as
it precludes fragmentation after the sender).

[toc] | [prev] | [next] | [standalone]

#3716

On Sat, 2016-04-23, Pascal Hambourg wrote:
> Popping mad a écrit :
>> 
>> "Fragments need to be reassembled before they reach the transport layer at 
>> the destination. Indeed, both TCP and UDP are expecting to receive 
>> complete, unfragmented segments from the network layer."
>> 
>> I thought that was wrong.
...
>> They have to be reassembled AT the destination, not before that.
>> 
>> Where am I confused, or is this in fact incorrect?
>
> IPv4 does not preclude reassembly before the destination.

True, but that doesn't mean early reassembly is a good idea.  I
suppose firewalls and load balancers do it to some extent, but it's
best implemented in the destination IP stack.  Especially since there
may not be a single router which all packets to one IP destination
pass through.

(Even better: don't generate any fragments ...)

/Jorgen

-- 
  // Jorgen Grahn <grahn@  Oo  o.   .     .
\X/     snipabacken.se>   O  o   .

[toc] | [prev] | [standalone]

Back to top | Article view | comp.os.linux.networking

csiph-web