In an inter-networking fragmentation scheme (i.e. end-to-end), AIUI assembly
only occurs at tranmission destination, even if the packets cross multiple
networks. This is in contrast to the intra-network fragmentation scheme,
which reassembles at every gateway.

My question relates to the way that the fragmentation occurs at each
gateway. Obviously when the message leaves the source station it is
fragmented in a standard manner. This involves splitting the data into
multiple packets (generally the size of the maximum transmission unit
[MTU]). The original data is split into packets that are equal to the MTU
minus the packet overheads (e.g. header etc). However, I'm not entirely sure
what happens when the message reaches the next gateway.

Am I correct in saying that the packet created at the source station is then
fragmented (if necessary), at the next gateway? So, say a 1000 octet packet
(60 overhead + 940 data) reaches the next gateway - would that individual
packet then be fragmented? So, then say the MTU of the next network is 800
octets, and the 1000 octet packet was received - would the 940 octets of
data be extracted, then split into two packets, one of 800 octets (data +
overhead), another with the leftover (remaining data + overhead)?

TIA.

In article <S5xSh.2804$vo2.2284@newsfe3-gui.ntli.net>,
"John E" <La-la-land@hotmail.com> wrote:

> In an inter-networking fragmentation scheme (i.e. end-to-end), AIUI assembly
> only occurs at tranmission destination, even if the packets cross multiple
> networks. This is in contrast to the intra-network fragmentation scheme,
> which reassembles at every gateway.

Your second scheme doesn't make any sense. By definition there are no
gateways involved in intra-network communications. In any case,
reassembly never occurs anywhere other than at the destination.

>
> My question relates to the way that the fragmentation occurs at each
> gateway. Obviously when the message leaves the source station it is
> fragmented in a standard manner. This involves splitting the data into
> multiple packets (generally the size of the maximum transmission unit
> [MTU]). The original data is split into packets that are equal to the MTU
> minus the packet overheads (e.g. header etc). However, I'm not entirely sure
> what happens when the message reaches the next gateway.
>
> Am I correct in saying that the packet created at the source station is then
> fragmented (if necessary), at the next gateway? So, say a 1000 octet packet
> (60 overhead + 940 data) reaches the next gateway - would that individual
> packet then be fragmented? So, then say the MTU of the next network is 800
> octets, and the 1000 octet packet was received - would the 940 octets of
> data be extracted, then split into two packets, one of 800 octets (data +
> overhead), another with the leftover (remaining data + overhead)?

Yes. Unless the packet has the Don't Fragment flag set, in which case
the router will drop the packet and send an ICMP Fragmentation Required
but DF Set message back to the origin.

--
Barry Margolin, barmar@alum.mit.edu
Arlington, MA
*** PLEASE post questions in newsgroups, not directly to me ***
*** PLEASE don't copy me on replies, I'll read them in the group ***

On Apr 9, 1:11 pm, "John E" <La-la-l...@hotmail.com> wrote:

> Am I correct in saying that the packet created at the source station is then
> fragmented (if necessary), at the next gateway? So, say a 1000 octet packet
> (60 overhead + 940 data) reaches the next gateway - would that individual
> packet then be fragmented? So, then say the MTU of the next network is 800
> octets, and the 1000 octet packet was received - would the 940 octets of
> data be extracted, then split into two packets, one of 800 octets (data +
> overhead), another with the leftover (remaining data + overhead)?

Yes. Fragments are fragmented precisely the same way non-fragments are
fragmented.

DS

Hello,

Barry Margolin a écrit :
>
> In any case,
> reassembly never occurs anywhere other than at the destination.

However it appears that some firewalls and NATs do packet reassembly
because they need it for stateful packet inspection.

[Is this huge crosspost necessary ? FU2 comp.protocols.tcp-ip]

Pascal Hambourg <boite-a-spam@plouf.fr.eu.org> schrieb:

> Hello,
>
> Barry Margolin a écrit :
>>
>> In any case,
>> reassembly never occurs anywhere other than at the destination.
>
> However it appears that some firewalls and NATs do packet reassembly
> because they need it for stateful packet inspection.
>
Yes and some just drop them. With VPN I often had this problem. When the
packets where fragmented the connection became bad with a lot of packet
lost, especially when you go over a dial-in. Mostly I could solve the
problem in resizing the MTU of the VPN-Client to 1300-1400. I'm not sure
why the packet get dropped or not reassembled and I really would like to
know what is the reason. I guess they get dropped because their origin
is not the VPN-Client.

Florian

On Apr 9, 3:11 pm, "John E" <La-la-l...@hotmail.com> wrote:
> In an inter-networking fragmentation scheme (i.e. end-to-end), AIUI assembly
> only occurs at tranmission destination, even if the packets cross multiple
> networks. This is in contrast to the intra-network fragmentation scheme,
> which reassembles at every gateway.
>
> My question relates to the way that the fragmentation occurs at each
> gateway. Obviously when the message leaves the source station it is
> fragmented in a standard manner. This involves splitting the data into
> multiple packets (generally the size of the maximum transmission unit
> [MTU]). The original data is split into packets that are equal to the MTU
> minus the packet overheads (e.g. header etc). However, I'm not entirely sure
> what happens when the message reaches the next gateway.
>
> Am I correct in saying that the packet created at the source station is then
> fragmented (if necessary), at the next gateway? So, say a 1000 octet packet
> (60 overhead + 940 data) reaches the next gateway - would that individual
> packet then be fragmented? So, then say the MTU of the next network is 800
> octets, and the 1000 octet packet was received - would the 940 octets of
> data be extracted, then split into two packets, one of 800 octets (data +
> overhead), another with the leftover (remaining data + overhead)?
>
> TIA.

Why do you want want fragmentation? Gateways are always pathways out
of your network. They are always routers.

On Tue, 10 Apr 2007 07:16:10 +0000, Florian Zschocke wrote:

>> However it appears that some firewalls and NATs do packet reassembly
>> because they need it for stateful packet inspection.
>>
> Yes and some just drop them. With VPN I often had this problem. When the
> packets where fragmented the connection became bad with a lot of packet
> lost, especially when you go over a dial-in. Mostly I could solve the
> problem in resizing the MTU of the VPN-Client to 1300-1400. I'm not sure
> why the packet get dropped or not reassembled and I really would like to
> know what is the reason. I guess they get dropped because their origin
> is not the VPN-Client.

The most common cause of this is a pmtud blackhole. Somewhere along the
path the mtu is smaller. Somewhere before this point the icmp-
fragmentation-needed packages are dropped.

HTH,
M4