L3 - Switching Flashcards
What is the purpose of ARP (Address Resolution Protocol)
It is a way for host to learn MAC address of other host
What are the steps of performing ARP?
1 - A host queries with an IP address, broadcasting the query to every other node on the network
2 - Host that has that IP address on the LAN will respond with appropriate MAC address to the requestor
3 - When host that issues a query receives reply, it builds an ARP table
4 - ARP table maps each IP address on LAN to the corresponding MAC address. Host can simply consult this table for future reference instead of broadcasting another ARP query.
What is a layer 2 datagram?
When host wants to send a packet to destination with a particular IP address, it takes IP packet and encapsulates it in an ethernet frame with corresponding destination MAC address. Essentially puts IP packet inside of the ethernet frame. Before sending IP packet with destination IP address, it first puts the packet inside a larger ethernet frame with own source MAC address and destination MAC address from local ARP table
What are two ways we can connect LANs?
1 - Hubs (these just re-broadcast to everybody)
2 - Switches (these can partition a LAN into multiple segments. Enforcing this isolation requires a switch table that maps destination MAC addresses to output ports)
What is a Learning Switch?
A Learning switch is the most common type of switch, and it maintains a table between destination addresses and output ports on the switch
How do we solve the problem of cycles in networks when flooding the network?
We form a spanning tree - a loop free topology that covers every node in the graph of the underlying physical topology.
What used to be the ‘rule of thumb’ for buffer sizing in routers?
2T*C
where T is the single trip time (in sec) and C is the capacity of the bottleneck link (in bits/sec). 2T*C is effectively the number of bits that could be outstanding on this path between the source and the data at any one time
What is the ‘new’ rule of thumb for buffer sizing, for when flows are desynchronized?
2T*C/root(n)
where T is the single trip time (in sec), C is the capacity of the bottleneck link (in bits/sec), and n is the number of unique congestion windows.
