[ns] Validation errors under linux (mandrake)

["Afrodite Sevasti" <sevastia@cti.gr>]

Hi!!

 this is to confirm a problem reported by "Marky S."
<markybeach@hotmail.com> on the 8th Dec. 2001 for validation failures of ns
2.1b8a on Mandrake Linux. I have the same problem exactly and I have tried
to describe it more thoroughly below. Needless to say, I've read the FAQ,
ns-problems page, and manual and I couldn't find the answer there. There are
some hints about Linux on Intel HW validation failures but nothing specific
and not specific fixes. The only specific suggestion for a patch to
TclAppInit.cc that I have found in the ns-users mailing list does not solve
the problem (see below).

I have already used diff to spot the differences that cause the validation
problem and all differences seem to be in arithmeting floating point
results. I have tried to validate with both the original distribution of ns
2.1b8a and the ns-2 floating point fix for Linux patch to TclAppInit.cc by
Tom Kelly (found in http://www-lce.eng.cam.ac.uk/~ctk21/code/) but without
luck.

If anyone has a solution to this, PLEASE let us know as soon as possible

 Problem description:
-----------------------------
Category:  Validation Failure
Package:   ns 2.1b8a
OS:        Mandrake Linux 8.1 kernel 2.4.8-26mdk
Environment Variables:
LD_LIBRARY_PATH=/usr/local/ns-allinone-2.1b8a/otcl-1.0a7:/usr/local/ns-allin
one-2.1b8/lib
TCL_LIBRARY=/usr/local/ns-allinone-2.1b8/lib/tcl8.3
TK_LIBRARY=

 -----------------------------
Description:

 Description of Problem:
After compiling ns-allinone-2.1b8 I get some validation errors. The
following tests differs from the reference output:

 ./test-all-tcp ./test-all-testReno ./test-all-newreno ./test-all-sack
./test-all-tcpOptions ./test-all-tcpReset ./test-all-tcp-init-win
./test-all-tcpVariants ./test-all-aimd ./test-all-rfc793edu
./test-all-rfc2581 ./test-all-ecn ./test-all-ecn-ack ./test-all-ecn-full
./test-all-red ./test-all-red-pd ./test-all-rio ./test-all-cbq
./test-all-schedule ./test-all-links
./test-all-oddBehavior

 How Easily Reproducible:
 Every time.

 Steps to Reproduce:
 1. Run ./validate

 Actual Results-Expected Results:
 For all tests failing, examination of contradicting results, by using diff
 presents differences in the floating point numbers. For example running
 "diff test-output-tcp/ecn.test test-output-tcp/ecn" gives the following
 differences (first 10 lines and last 8 lines)

 > 0.006 0
 > 0.006 0
 > 0.06744 0.01
 > 0.06744 0.01
 > 0.06844 0.02
 > 0.07744 0.02
 > 0.12888 0.03
 > 0.12888 0.03
 > 0.12988 0.04
 > 0.13888 0.04
 .....
 > 5.18176 1.8
 > 5.71176 1.04
 > 6.73176 1.28
 > 3.24176 1.05
 > 5.18176 1.8
 > 5.71176 1.04
 > 6.73176 1.28
 > 7.82176 1.53

 Additional Information:
1. Trials with both the original distribution of ns 2.1b8a and the ns-2
 floating point fix for Linux patch by Tom Kelly (found in
 http://www-lce.eng.cam.ac.uk/~ctk21/code/) failed
 2. My 'validate.log' file is exactly the same as the one in
 http://www.iam.unibe.ch/~studer/nstestoutput.txt


 Many thanks for anyone who will try to help
Afrodite