CS625 – Advanced Computer Network
Instructor – Bhaskaran Raman
Lecture 16 ( Sep 09, 2003)
Scribe – Rajiv Kumar Ranjan (Y3111037)

Asymmetric networks

The common asymmetries exist in a network due to :

New network technologies provide much larger downstream bandwidth (towards the user) than upstream bandwidth (away from the user). The asymmetry in bandwidth could range from a factor of 10 to 1,000.

In certain wireless networks, the underlying media access control (MAC) protocol often results in a significantly larger one-way latency from a base station to a mobile station than in the reverse direction. In other networks there is a significant delay when a node switches from sending to receiving mode or vice versa.

Packet errors:
The incidence of packet errors could be much greater in the upstream direction than in the downstream direction. This could be inherent in the network technology or the result of distinct upstream and downstream technologies .

These asymmetries could have an adverse effect on TCP performance due to several reasons: the limited upstream bandwidth could throttle the flow of TCP acknowledgment packets and disrupt TCP's ack clocking mechanism, the asymmetry in latency could cause large variations in TCP's roundtrip time estimate, etc.

We will concentrate mainly over the asymmetry caused by band-width, because it is much common than other ones.

Band-width Asymmetry:
Asymmetric links are characterised by having different bandwidths for each direction of the communication. There are many TCP applications such as Web access or FTP in the Internet that take advantage of this fact, because they involve a substantially larger flow of data in the forward direction (from host to client) than in the reverse direction ( from client to host). These applications are based on the TCP.

TCP performance can suffer an important degradation when data is transmitted through asymmetric links.When a connection is established, packets are sent through the forward channel and acks through the reverse channel. If it is generated one ack for each received packet, acks are generated faster than the link rate.As the TCP behaviour is self-clocking, property according to which the transmission of packets at the sender is made after the reception of acks, the forward data transfer can be slow down.

Thus, asymmetry in band-width might cause

( Normalised bandwidth ratio:
The Normalised bandwidth ratio, k, between the forward anfd reverse paths is the ratio of the raw bandwidth divided by the packet sizes used in the two directions.
If forward channel bandwidth = 10Mbps , Packet size = 1000 bytes,
    reverse channel bandwidth = 10Kbps and ack size = 40 bytes,
then, normalised bandwidth ratio = (10Mbps/1000B)/(10Kbps/40B)= 40. )

The possible solutions are:

Ack congestion control :
This is basically an end-to-end solution done at the receiver. If TCP receiver can learn of Ack loss, then it can send Acks at a slower rate. For this, the mechanism employs a dynamically varying delayed-ack factor ( this factor represents the no. of packets received for which single ack is sent ) based on the congestion of acks in the reverse link. Although this mechanism is effective to control the congestion of the reverse link, its effectineness reduces in the presence of packet losses due to errors.

Ack filtering mechanism:
It is based on controlling the amount of acks enqueued at the reverse link router. When a new ack is received, some fraction of the acks ( possibly all but one) belonging to the same connection is removed.Therefore, the frequency of the acks preventing the slow down of the forward data transfer.Its effectiveness against congestion or losses, because it acts directly where it takes place. Here no per-flow state is required in router.

Sender Adaption:
In this approach, the sender sends according to the number of segments acknowledged and not based on the number of Acks. i.e. the window growth is tied to amount of data acked rather than the number of acks received. So, potentially large bursts are broken up into smaller ones.

Acks-first Scheduling:
Router can schedule Acks with priority. Here, again, no per-flow state is required.

Mobile IP

Mobile IP is an open standard, defined by the Internet Engineering Task Force (IETF) RFC 2002, that allows users to keep the same IP address,  stay connected, and maintain ongoing applications while roaming between IP networks. Mobile IP is scalable for the Internet because it is based on IP -- any media that can support IP can support Mobile IP.

Before discussing the various issues of mobile IP, we first address the mobility problem.

The Mobility Problem:

"If a host moves only locally, there may be no problem if the local subnet is flat addressed. But, Longer distance moves require global routing updates, which is obviously impractical in a large network even for small no. of moving hosts."

Although Hierarchical, topologically- based addressing ( in which at least part of the address identifies approximate location ) is fine for scaling a network consisting of nodes whose network attachment points don't change, it is not suitable for nodes which are movable ones (i.e. Mobile nodes). The reason is that the routers only need to know the way to each network, but if a host moves, its packets will still go to its home!

The above problem arises due to the fact that Internet routing is titghtly coupled with addressing and the usual addressing couples the identity and location of the nodes. So, we need a way to seperate identity from location. For this, we introduce a level of indirection on the addressing scheme, which forms the basis of the Mobile IP. 

Mobile IP works in the following way:

  1. Leave the hierarchical network intact

  2. Craete special entities that "own" the mobile host address

  3. Mobile hosts report their locations (register) with the stationary entities

  4. Traffic to the mobile Hosts is relayed by the stationary entities.


The Basic Architecture:

Each mobile host (MH) has a home agent (HA) that acts as a location registry, maintaining the binding from the MH's home address and its foreign care-of-address. Each time the MH moves, this binding is updated. Correspondent hosts (CH's) address messages with the MH's home address; the HA forwards this message to the MH via IP tunneling or encapsulation. MH's gain their care-of-addresses by contacting a foreign agent; the foreign agent assigns the care-of-address and updates the MH's binding by contacting the HA. This updation requires some authentication informations. The whole pocess is explained below.

The Mobile Host is a device such as a personal digital assistant, or laptop whose software enables network roaming capabilities.

The Home Agent is a router on the home network serving as the anchor point for communication with the Mobile host; it tunnels packets from a device on the Internet, called a Correspondent Host, to the roaming Mobile Host. (A tunnel is established between the Home Agent and a reachable point for the Mobile Host in the foreign network.)

The Foreign Agent is a router that may function as the point of attachment for the Mobile Host when it roams to a foreign network, delivering packets from the Home Agent to the Mobile Host.

The care-of address is the termination point of the tunnel toward the Mobile Host when it is on a foreign network. The Home Agent maintains an association between the home IP address of the Mobile Host and its care-of address, which is the current location of the Mobile Host on the foreign or visited netwok.

The basic Architecture of Mobile-IP is shown below:


Working of Mobile IP:

The Mobile IP process has three main phases, which are discussed below:

Agent Discovery: 
A Mobile Host discovers its Foreign and Home Agents during agent discovery. During the agent discovery phase, the Home Agent and Foreign Agent advertise their services on the network. The Mobile Host listens to these advertisements to determine if it is connected to its home network or foreign network. If a Mobile Host determines that it is connected to a foreign network, it acquires a care-of address which is provided by the foreign agent. When the Mobile Host hears a Foreign Agent advertisement and detects that it has moved outside of its home network, it begins registration.

e Mobile Host sends the registration request to its Home Agent. The Home Agent checks the validity of the registration request, which includes the care-of-address and authentication of the Mobile Host. If the registration request is valid, the Home Agent creates a mobility binding (an association of the Mobile Host with its care-of address), a tunnel to the care-of address, and a routing entry for forwarding packets through the tunnel.
Thus, a successful Mobile IP registration sets up the routing mechanism for transporting packets to and from the Mobile Host as it roams.

The Mobile Host sends packets using its home IP address, effectively maintaining the appearance that it is always on its home network. Even while the Mobile Host is roaming on foreign networks, its movements are transparent to correspondent Hosts. Data packets addressed to the Mobile Host are routed to its home network, where the Home Agent now intercepts and tunnels them to the care-of address toward the Mobile Host. Tunneling has two primary functions: encapsulation of the data packet to reach the tunnel endpoint, and decapsulation when the packet is delivered at that endpoint. The default tunnel mode is IP Encapsulation within IP Encapsulation.
Finally, the Mobile Host sends packets directly to their final destination, i.e. the Correspondent Host.

The following figure summarises the above process:


MH: Mobile Host
CH: Correspondent Host
HA: Home Agent
FA: Foreign Agent