Re: about TCP initial window test suite

[Sally Floyd <floyd@aciri.org>]

>   Can anyone help me with this particular test suite named
>"test-suite-tcp-init-win.tcl" which is in the ns/tcl/test directory or
>can u suggest someone who can help me with the same??

I wrote that test suite, and the tests are documented in the
code "tcl/test/test-suite-tcp-init-win.tcl" and in the
accompanying file "tcl/test/test-suite-tcp-init-win.txt",
which I have also appended below.

In the simulations, the "receivers advertised window", or the
upper bound on the congestion window, is set with the
command "$tcp1 set window_ 40", and the initial window,
when non-standard, is set with the command 
"$tcp1 set windowInit_ 6".

I am in Oslo at IETF at the moment, and will rely of someone
else to answer your questions about the window size and
calculating delay.

- Sally

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

test-suite-tcp-init-win.txt:

The tests of options for TCP's initial window can be tested
with the command "./test-all-tcp-init-win" in the directory
tcl/test.

These tests verify the following options added to the
implementations of one-way TCP.  The proposal in the
internet draft "Increasing TCP's Initial Window" can be
modeled by setting syn_ and delay_growth_ to true, and
setting windowInitOption_ to 2.

For these parameters, the model implemented in the simulator for
one-way TCP is as follows:  the sender sends a 40-byte data packet
(modeling the SYN), the receiver sends an ACK, and the sender then
sets its initial window to either two, three, or four segments,
depending on the segment size.  If the first packet (i.e., the SYN)
is dropped, then the initial window is started at one segment.

Agent/TCP, syn_:

For the default, with "syn_" set to false, TCP does not model the
initial SYN/ACK exchange.  With "syn_" set to true, TCP does model
the initial SYN/ACK exchange, in that for "syn_" set to true, the
initial packet is of size tcpip_base_hdr_size_ (40 bytes).

Agent/TCP, delay_growth_:

For the default, with delay_growth_ set to false, the initial window
applies to the first packet.  With delay_growth_ set to true,
the initial congestion window only applies after a single packet
has been sent and acknowledged.

Agent/TCP, windowInitOption_:

For the default, with "windowInitOption_" set to 1, the TCP initial
window is set to the number of packets specified in "windowInit_",
which itself is set by default to one packet.  For "windowInitOption_"
set to 2, the TCP initial window is set to the number of packets
specified in internet draft draft-floyd-incr-init-win-01.txt on
"Increasing TCP's Initial Window".  That is, for a packet size of
at most 1095 bytes, the initial window is set to four packets.
For a packet size between 1095 and 2190 bytes, the initial window
is set to three packets, and for a packet size of 2190 bytes or
more, the initial window is set to two packets.

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

The first tests, tahoe1 to tahoe4, test the initial window options
with Tahoe TCP.  

The three connections in the tahoe1 test all use "syn_" and
"delay_growth_" set to true, and "windowInitOption_" set to 2,
for the initial window specified in the internet draft.
The packet sizes for these three connections are 1000 bytes, 1500 bytes,
and 4000 bytes, respectively.

The three connections in the tahoe2 test use various combinations
of fixed initial windows, with "windowInitOption_" set to 1.
Two of the connections use an initial window of six packets, and
one uses an initial window of one packet.

The connection in the tahoe3 test shows that with "windowInitOption_"
set to 2, when the initial (SYN) packet is dropped, the larger initial
window is not used.

"Init4" shows two connections with 1000-byte packets using the initial
window specified in the internet draft.  The test shows that after a
Fast Retransmit, TCP slow-starts from an initial window of one packet.

These four tests are then repeated for Reno, NewReno, and SACK TCP.