readme.notes

柯老师网站上找到的

Test Scenarios:

---------------------------------------------------

Look at time-sequence plot, infer TCP characteristics.
One-way bulk data transfer.

1.  Send N data packets in all, no losses.  Check for Delayed Acks
	(Sec. 3.7 of ID), initial windows, slow start (Sec. 3.1 of ID),
	receivers advertised window.  Check for FIN exchange (Sec. 3.8 of ID).

2.  Drop the SYN packet.  Check initial RTO.

3.  Drop the first data packet.  Check RTO after one RTT measurement.

3A.  Repeat 3, also dropping the first retransmitted packet.  Check for
	retransmit timer backoff.

3B:  Repeat 3, dropping every retransmitted packet.  Check for when
	connection terminates.

4.  Drop a single data packet 20 or more data packets have been sent.
	Check for Fast Retransmit instead of Slow-Start.
	Check for failure of receiver to retain above-sequence data
	(Sec. 3.4 of ID).

4A.  Repeat 4, also dropping the first retransmitted packet.  Check for more
	or less aggressive implementations of SACK (depending on whether it
	waits for a retransmit timer.)

5.  Drop an entire congestion window of packets, forcing a
	Retransmit Timeout.  (How to do this:  Drop packet N,
	and drop all succeeding data packets until you see a
	data packet with retransmitted data.) Check for Slow-start
	(Sec. 3.2 of ID), check RTO and make inferences about TCP
	clock granularity.

6.  Drop a single data packet after many packets have gone by.
	Let K dup acks go by in the other direction, and then drop
	all succeeding dup acks until a new ack is seen.  Repeat
	for K=1,2,3,4.  Check for dup ack threshold, any responses
	to one or two dup acks.

7.  Drop K consecutive packets after many packets have gone by.  Check for
	Tahoe (slow-start after retransmit timeout or Fast Recovery), Reno 
	(might or might not have to wait for a Retransmit Timeout,
	depending both on K and on the number of dup acks), NewReno,
	SACK (no unnecessary retransmission of packets even for K=2)

8.  Multiple Fast Retransmits.  Reproduce Figure 11 of the Simulator Tests document.

9.  Drop first and last packet from a congestion window.  (Figure 3 of
	"Simulation-based Comparisons of Tahoe, Reno, and SACK TCP").  Look
	for Tahoe vs. Reno vs. New-Reno or SACK.

10.  Drop packets N, N+2, N+4, N+6, for large N.  Check for retransmit timeout
	for Reno, not for Tahoe, NewReno, or SACK.  (Figure 4 of
	" Simulation-based Comparisons of Tahoe, Reno, and SACK TCP".)

Sec. 3.9 of ID: a large, interrupted bulk transfer.
Sec. 3.10 of ID: a transfer in which the receiving application terminates abnormally.
Tests of SACK vs. FACK.
Two-way TCP
Initial window on restart.

---------------------------------------------------
Uses:
Testing researchers' experimental implementations - SACK, ECN, NewReno.
Testing production TCP implementations.
Testing beta TCP implementations (e.g., Windows 98, NT 5, etc.)
Testing a TCP connection when you don't have access to the remote
  end, and are curious about what generation of TCP they are using.
Plug into tcpanaly: