Side_1_360

• Sender adaptation; when an ACK message
  acknowledges a number of data segments, one
  could equate it to the same number of ACK
  messages. Thus, the congestion window at
  the sender side can grow correspondingly.
  Making sure that the number of segments
  acknowledged and not the number of ACK
  messages is used would also be in line with
  the situation on the forward direction.


• Reconstructing ACK messages; in case sender
  adaptation is not used, the original sender side
  of the constrained link (forward direction) can
  examine the ACK messages and reconstruct
  any intermediate ACK messages not seen.
  This could be done by generating the ACK
  messages and putting them on the link evenly
  distributed in time.

A more generic technique of the latter is to intro-

duce an ACK compaction and an ACK expan-

sion in the network, see Figure 14. The com-

paction would remove any “unnecessary” ACKs

and the expansion would reconstruct the same

ACKs, making it transparent for the end-sys-

tems. However, additional protocol mechanisms

have to be introduced to enable this.

6 Addressing and Routing

6.1 Addressing and Identifiers

A nameidentifies whatan object is, an add-

ressidentifies whereit is, a routetells howto

get there, a pathsays which sequence of steps

to traverse, and a linkwould be a stepin the

path.

Referring to the configuration depicted in Figure

15 a major issue is how the different port num-

bers are allocated. In principle there are two

ways this can be done; i) universal assignment;

and ii) dynamic binding. In the former, a cen-

tralised authority is typically used to assign port

numbers and publish the results to the hosts

(could well be done hierarchically). In the latter,

port numbers are assigned when needed. This

implies that each program that needs a port num-

ber is assigned one on demand. In order to know

the port number on a remote host, an enquiry has

to be sent to that host, which replies with the

proper port number to use.

Typically a combination of the two ways of

assigning port numbers has been chosen; some

numbers are fixed while others are used dynami-

cally. The ports refer to identifiers used on the

transport layer (TCP/UDP). The port identifiers

are included in the UDP/TCP headers.

Referring to identifiers used on the IP layer, a

Domain Name System (DNS) is used to translate

between more high-level names and IP add-

resses. For instance, the name viking.telenor.com

could translate into a 32 bit IP version 4 address

(and the other way around). Such a naming

scheme would then be used to assign names

throughout the IP-based networks. It also pro-

vides a large-scale example of the client-server

concept as a DNS server would be enquired in

order to make a translation between the name

and the address, see examples in Figure 16.

In principle, the resolution algorithm used for

the translation proceeds from the top (top-level

domain) and continuing down. There are two

ways of using the domain name system: i) by

asking the name servers one by one until the

resolution is complete; or ii) by asking a name

server to do the complete resolution (lower

Figure 14 Introducing ACK
compaction and expansion
nodes (note these might be the
receiver and the sender nodes,
respectively)

ACK expansion node ACK compaction node

sender receiver

ACK

data

ACK

data

ACK

data

ACK

data

ACK

data

ACK

data

constrained reverse link

e.g. A socket

A port identity

A IP address

Application

Transport

IP

Network

Interface

IP

routing/forwarding

Network Interface

Application

Transport

IP

Network

Interface

e.g. B socket

B port identity

B IP address

Host A Host B

User data Transport layer header IP packet header Link level header

Figure 15 Hierarchy of
functionality/protocols