Comp 388/488: Advanced TCP/IP Networks

Final exam study guide

Here is a brief outline of the material that may be covered:

You will still need to know the basics of congestion-window management
of Tahoe/Reno/SACK TCP. Other than that, the topics covered on the final
will be things covered in class since spring break, plus the BGP notes
(most but not all of which was in fact covered since spring break). 

Study materials:
        notes
        bgp notes
        fair queuing notes
        selected papers on my web page (but for every topic that may
           appear on the exam there is also fairly detailed coverage
           in one of the notes files)

Multicast: Section 4.4 of Peterson & Davie.
        Link-state multicast (easiest case)
        Distance-Vector multicast (trickier)
        Protocol-independent multicast (sparse mode)

TCP and fairness
        consequences of longer RTT for slower bandwidth increase
        How does Reno TCP bias against long-RTT connections?
        How does Reno TCP bias against multiple congested links?
        queue size and bandwidth share

RED gateways: P&D 6.4.2
        What problem is it that these attempt to address?
        What are some of the fairness issues regarding DECbit,
        as a per-connection queue management technique?
        
Bottleneck link
        Packetpair as a technique for probing the bottleneck link
        Link bandwidth can be the actual bitrate, eg 56Kbps, or it
        can be the average fraction of bandwidth available to one 
        connection; eg a 1.5mbit line shared among ten connections
        means that each connection gets 0.15mbit.

TCP vegas, P&D 6.4.3
        basic ideas
        what Vegas uses as congestion indicators; how Vegas responds
        costs of errors/misinterpretation in the TCP Vegas measurements
        How Vegas competes with Reno.
        
TCP models
        What assumptions does the network "fluid" model (eg in Mo & Walrand's
        paper) make? 
        How do tiny packets affect transfer A--R--B, from A to B via router R?
        What happens to "bandwidth delay"?
        
Integrated Services (RSVP; P&D 6.5.2) v Differentiated Services (6.5.3)
Why RSVP is "expensive"
Why RSVP does a good job of supporting multicast
Token bucket filters (Differentiated Services)
        What token-bucket bucket depth is needed to send the following
                Average rate: 1 packet/sec
                Each 8 seconds:
                        T=3:    3 packets
                        T=8:    5 packets
                        
Issues with figuring out the path
"Premium bit" in Differentiated-Services headers

BGP (see notes): 
        Examples of BGP policies
              customer/provider/sibling/peer
        unstable BGP relationships

Fair Queuing: basic rules (P&D 6.2.2)
        How Fair Queuing changes the rules re per-connection queue management
        Simulated bit-by-bit round-robin: measuring time with a clock that
        runs at 1/N normal rate when N>1 queues are nonempty.

Realtime/Quality_of_service issues
        service categories
        typical requirements
        
RTP/RTCP: realtime protocol (P&D 9.3.2)
        Consequences of no more "ACK self-clocking"
        Application-level control over framing & transmission
        Living without ACK self-clocking
        RTCP "5% rule": may limit frequency of "ACKs"
        Comparison of RTCP receiver reports with TCP-style ACKs
        Possibility of use of jitter information
        How RTP competes with TCP

TCP-Friendly Rate Control (TFRC)
        basic operation
        how equation-based rate control works
        how we get from a window size to *rate* control
        What happens if congestion suddenly gets worse
        What happens if (old rate) > (new avail bandwidth)
        How AI/MD works with different decrease factors / increment amounts