Lecture 2 Summary

CS 4700 / CS 5700
Network Fundamentals
Lecture 7.5: Summary from Lecture 2
Revised 1/25/2014
Project 1 in, Project 2 out (Friday)
2


Project 1 will be graded next week
Project 2 out Friday
Firewall vs closed port
3

Inside rack
 Port
27993
 OK
 Port
27992
 ICMP
type 3 (destination unreachable)
Firewall vs closed port
4

Outside rack
 Port
27993
 OK
 Port
27992
 Blocked
 Timeout
(no response)
Summary from last class
5


C Sockets
Physical layer
 Move
bits around
 Synchronize to determine when bits start/end

Link Layer
 Framing
 Error
detection
Today’s class
6

Paper reviews
 Spanning
Tree
 OSPF in operational networks

Network layer
 Addressing
 Routing
 Packet


delivery
Intradomain routing
Next week: Interdomain routing
Perlman paper
7

This paper won her a SIGCOMM award in 2010

Key take-aways
 Not
reasonable to assume topologies are loop free, and we
need a way to route between LANs automatically
 Low memory consumption
 Scalable bandwidth consumption
 Converges quickly
 Tunable
Perlman paper (2)
8

Implementation details
 Timers
everywhere
 How
else do we know when a link is down?
 Also can prevent transient loops (hold down)
 Loop
 This
detection/avoidance
comes up often in networking
 Deterministic
behavior
 Allows
operators to reason about changes
 Very hard to do in wide area!

Compare this with BGP when we get to it
Shaikh paper
9

Key take-aways
 Understanding
 Especially
 Best
the state of large-scale networks is hard
when they are self-managing
way to monitor is to participate in the protocol
 Real
time is critical, so separation of functionality is critical
 Offline vs online analysis
 Must not adversely affect the network
 Modeling
the network allows us to identify anomalies
 Flaps
 Message
 Etc
storms
Shaikh paper (2)
10

Networking and operators
 Operators
are extremely cautious people
 Tend to lack good tools for understanding the network

Routers are not perfect
 Router
bugs, flapping due to load, improper timers
 Generally hard to detect if you’re not looking for it

Humans are definitely not perfect
 Bad
configs, e.g.