DHCP
- assigns IP to devices on address changes, every time you connect to a network you get a new one
- there is a lease time, and if you leave after that lease time the ip will be reclaimed and can be reassigned to another computer
- a newly booted hosts sends a DHCPDISCOVER message to the broadcast address.
- Because having a DHCP server for every business or floor of an administration is laborious, you can use relays that sole job is to relay discover messages to DHCP server
- DHCP is derived from BOOTP
Error Reporting (ICMP)
- defines a collection of error messages that are sent back to the source host whenever a router or host is unable to process an IP datagram successfully.
- along with defining many error messages that can tell hosts what the problem is, it also provides the debugging tools of traceroute and ping
Address Resolution ARP
- each host has a table that translates MAC addresses into IP addresses
- entries time out roughly every 15 minutes.
- the currently stored map is known as ARP cache or table
- ARP sends a broadcast request for an IP and if it matches the host that received it, the receiver replies with the corresponding MAC address.
- if the IP is already in the table, ARP refreshes timer
UDP
- does not guarantee delivery of packet to the destination. uses “best effort”
Subnetting
- network numbers managed by ICANN
TCP
TCP Congestion Control
Overall principle is for each node to gauge how much traffic is in the network and try to not surpass the network’s limit
- uses the arrivals of ACKs to count the number of packets it has sent vs received to keep it constant
- This “self clocking” paces the packets
- How do we know the networks limit? since it is always changing
- Additive increase, multiplicative decrease
- every time a packet is successful, it adds 1 to its congestion window size count
- every time a packet is lost, the congestion window is 1/2
- slow start
- when a packet is successful, the congestion window is doubled
- usually used at start of connection when no data is known of limit
- after some point it returns to a linear growth and retraction
- Fast retransmit and fast recovery
- if at any point the sender notices three ACKs of the same packet, it resends, even if that is before the RTT
- Additive increase, multiplicative decrease
- can have problem over wireless connections where bit errors are detected as packet losses, which causes the TCP to slow to a crawl unnecessarily
Quality of Service
- as data sizes increase we need more data to be able to transfer at faster rates
- time sensitive applications are call real-time applications
- real-time apps need assurance from the network that data is going to arrive on time
- A network that can provide different levels of service (real time vs best effort) is said to support QoS
- non-real time also called traditional data apps or elastic
- email, web browsing
- intolerant real time apps can get guaranteed serrvice
routing
The purpose of routing is to create a routing table
- routing can be boiled down to a graph theory problem
- the caveats being that some edges can disappear and new ones can be created
- as well as the weights of the edged changing over time
Link State
- each node is able to find the distance to its neighbors, and their state.
- Link state relies on eventually every node being able to create a complete map of the network
- each node floods the network with its id, list of connected neighbors, sequence no, and TTL
- uses periodic hello packets to lets its neighbors know it is still active.
- LSP packets carry TTL, meaning old state information is automatically ignored from the network, preventing false images of the network being created
- Once the network image “converges” the router can perform dijkstra’s algorithm to find the shortest path to all nodes.
- stabalizes quickly from removed nodes, and doesn’t generate an abundance of traffic
- requires high storage on each node to keep track of the topology
OSPF
- Most widely used Link state protocol
- OSPF has an authentication message to ensure the routers getting and sending information are to be trusted
- allows somain to be partitions into areas, this means a router doesn’t need to know how to reach nodes outside its area, only how to reach said area
- OSPF allows different paths to have the same weight and thus distribute the load across said paths.
Distance vector routing
- each node constructs an array / vector containing the distances to all other nodes it knows of, and distributes that to its neighbors.
- a down node is given infinite distance
- similar to the bellman ford / floys warshall algorithms that we discussed in data structures.
- each node automatically sends an update periodically “keep alive”
- triggered update happens when a node notices a node has been disconnected
Routing Information Protocol
- RIP is the most common used protocol implementation of Distance vector.
- it was distributed iwth BSD version of unix.
- send a keep alive every 30 seconds
- RIP supports subnet masks
- limited to hop distance of 15, and thus considers 16+ hops to be non-reachable
BGP
- The Exterior Gateway routing protocol
- BGP is a protocol used between ASes
- BGP is a form of Distance vector, but rather than using minimum distance, policy is used to pick the route.
- BGP also has to keep track of the specific route that it took along with the cost associated with it, something RIP didn’t have to do intradomain
IPv6
Virtual Circuit
- provides a logical point-to-point connection along a leased connection line
- part of the path is stored in the routing table so that it doesn’t have to searched for again every time we want to send a packet.
- each packet carries an identifies telling which cirtual circuit it belongs to, and upon arrival the vc dissipates