CS625 -- Lecture 4 (04 Aug 2003)

CS625 - Advanced Computer Networks
Instructor - Bhaskaran Raman
Lecture 04 - 4th August 2003
Scribe - Paul Ipe (Y0224)

Overview of Lecture

Internet Routing Architecture
Border Gateway Protocol

Routing Protocols (recap)

Link-State
Each router is expected to maintain atleast a partial map of the network and when any link changes state the network is flooded with the notification. All the routers note this change and recompute their routes accordingly. This is computation intensive yet simple.
Distance-Vector
Each router is expected to share with its neighbours its routing tables. Receiving routers can choose neighbours advertising lowest cost to a destination and add this entry to its routing tables.
Path-Vector
Cost to all nodes in the graph is calculated.
Entire path information is passed.
This avoids the problem of local looping and count to infinity which can arise with the link-state and distance-vector routing protocols.

Other types of routing such as hierarchical routing also exist where the path from source to destination follows a fixed hierarchy.

Address and Network Mask

Address Classes

Class A - For large networks

0 <7 bits of the network address> <24 bits of host address>
Class B - For medium networks

1 0 <14 bits of the network address> <16 bits of host address>
Class C - For small networks

1 1 0 <21 bits of the network address> <8 bits of host address>

Classless Addresses - When it is not defined as to which class a particular IP address belongs (A,B,C) then it is termed as a classless address. This can help solve the problem of too many hosts or too little hosts in a particular network.

A Network mask is used to determine the network number and host number of a particular IP address. This is done by a bitwise AND of the address and the mask.

Internet Service Providers

ISPs provide you with connectivity to the internet. For example, upon dial-up you can be given a dynamic IP address. But how do ISPs coordinate amongst themselves to provide connectivity to clients. One solution is peering.

There are in general two types of relations between Autonomous Systems

Peer to Peer - The business relationship whereby ISPs reciprocally provide to each other connectivity to each others' transit customers.
Transit - The business relationship whereby one ISP provides (usually sells) access to all desinations in its routing table.

Peering is a non-transitive relationship between ASes. An advantage of peering is that the customer does not have to buy bandwidth during transit. This cuts the cost to the AS or ISP. This also helps lower Inter-Autonomous System traffic latency.
As this is primarily a business model there can be disadvantages such as traffic assymetry (ratio of traffic out to traffic in) and no guarantees on quality of service provided (If a router goes down in AS-1 which peers with AS-2, AS-2 can give no guarantee to its clients when the router would be back up). These are details that can be managed by various business clauses.

Border Gateway Protocol

An Autonomous System is a group of networks using a single IGP (Internal Gateway Protocol) for routing information.
BGP is a path-vector based protocol. It is an EGP (External Gateway Protocol) -- it works between various ASes. It is interesting to note that the control path is opposite to that of the data path. BGP Router-1 gives BGP Router-2 information about the nodes that it can access. The flow of this control information is from 1 to 2. Now for data to go to any node accessible by 1 it would flow from 2 to 1.
BGP is implemented on top of TCP! - surprising as it is a routing protocol and should be implemented at the IP layer which is below the TCP layer. This is possible as BGP Routers between autonomous systems are generally neighbouring and a route already exists (hence doesnt have to be found). Put in another way, local reachability is implemented at the IP layer and the two BGP Routers can talk with each other as they are neighbouring. Now BGP packets are transmitted between them and using this data the routers can now implement global reachability between the two autonomous systems.
As autonomous systems have border routers that 'speak' BGP it is possible to implement policy based routing.

BGP Messages

OPEN - for initialization, exchange of AS numbers.
NOTIFICATION - before closing a connection.
KEEP-ALIVE - periodic heart beat to keep BGP session alive.
UPDATE - main routing messages, route announcements and withdrawals.

OPEN, NOTIFICATION and KEEP-ALIVE are required as BGP is required to explicitly withdraw a route that is announced unlike soft-state protocols where the state can get timed-out (eg. link-state protocol).

UPDATE Messages (BGP Path Attributes)

ORIGIN - How the route to the destination was learnt.
- IGP - it is within the AS
- iBGP - from another BGP Router within the same AS
- eBGP - from some other BGP Router in a different AS
AS_PATH - the whole path to the destination (to implement path-vector protocol)
NEXT_HOP - on the reverse path (the data path)
MULTI-EXIT-DISCRIMINATOR (MED) - When two autonomous systems are connected to each other by more than one link, the AS can assign weightages to the different links as it sees fit depending on the path.
LOCAL_PREF - If an AS has two or more routes to a particular destination then it can specify a preference as to which path it prefers (this is helpful in policy based routing).

Route Selection (Order of rules the router follows to determine along which path to send a packet)

LOCAL_PREF
AS_PATH
MED
eBGP > iBGP
IGP cost

The above pretty much covers what was done in class (Turns out not much was done as we had to discuss projects and a time-table change also). It would be advisable to also go through the assigned reading for lecture-04 which is put up on the course website.