2023-11-02
Summary
keywords forwarding
TODO
HW
Exercise*
Next time
Network Layer: data plane
Chapter 4 (data plane): forwarding, network layer model, how switches & router works, addressing, generalized forwarding, NAT, middleboxes Chapter 5 (control plane):
What is inside a router input ports, switching, output ports,
IP addresses.
SDN : Software Defined Network. Openflow
At network layer your header is called datagram.
Forwarding
Access point router will examine on network layer header. and moves the datagram from input port to output port.
Network switches
Two key network payer functions
1) forwarding
process of getting through an Interchange
2) routing
process of finding the trip routes.
Data plane
- local, per-router function
- determines how datagram arriving at input port is forwarded to output port
Control plane
- network wide logic.
- determines the forwarding table.
- two ways
- traditional vs. SDN
- in traditional way, the cost between routers are shared among routers.
- In SDN, the Control ageents inside routers share information only with distributed controllers. routers don't need to talk each other.
What is the benefit of SDN?
- easier for embedding new features.
Network Service model
What do we require to the service model of a network?
#todo : difference between network architecture, service
Internet
- "Internet" model is best effort service model of this time.
- It doesn't do any guarantee of Quality of Service (throughput, loss, ordering, delay).
- Those properties are guaranteed on TCP, a transport layer.
- Core is simple, and all the complicated logic is done at the edge.
Other Network Service models
- ATM architecture,,,,,
- They are not popular because of too much overhead.
- There are some other service models with Internet architechture. to implement, the routers configuration is needed.
Why is best-effort service model is popular?
- simplicity of mechanism
- easy to extend the network.
- we have enough bandwidth for current traffic, and good enough performance.
What is inside the router?
router has input ports, high-speed switching fabric, routing processor, output ports. #todo : add image of the router
in high-speed switching fabric, all logic is embedded in hardware level.
input port
- read the 0,1,0,1 coming from the line termination.
- comprehend the link layer protocol
- lookup in input port memory
- and forward to the output port through the right switch fabric This is called decentralized switching. This forwarding should be completed at line speed. Otherwise, queueing happens inside.
- Destination based forwarding : forward only based on destination IP address (prefix matching). Simple and affordable.
- generalized forwarding(SDN) : it compares many other header fields for priority, for firewall, ...complicated and expensive.
Destination-based forwarding
#todo : add example forwarding images.
It is range-based. It is based on the longest prefix matching.
Longest prefix matching
to catch up the line speed, specialized hardware, named TCAMs, is used. TCAM is a memory space with come computational circuitry. can do in one clock cycle. TCAMs is content addressable.
</prefix matching>
</switching>
Switching Fabrics
switch rate : N input ports, S line speed = N*S switch rate Three major types of switching fabrics
- memory
- bus
- interconnected fabric
- Memory
- speed limited by memory bandwidth.
- 2 bus crossings per datagram.