gpsmap: "Floating point exception"

layer2 · 03-01-2006

I've been unable to build any interpolated network power maps using gpsmap (from Kismet) version v2005.08.R1. ImageMagick is at 6.2.2. I've experienced the same floating point errors under Auditor, Fedora Core 4, RHEL3 and RHEL4. Any thoughts or advice appreciated.

Thanks...

$ gpsmap -D -z 4 -e -o out.png -f 00:0A:00:00:00:00 -i -p *.gps

... snip ...

Converting map to greyscale.
Calculating network coordinates and statistics...
Processing 1 raw networks.
Assigning network colors...
Plotting 1 networks...
Drawing network power interpolations...
Interpolating power into graph points.
Thread 0: crunching interpolation image line 0
Thread 1: crunching interpolation image line 5
Thread 2: crunching interpolation image line 10
Thread 3: crunching interpolation image line 15
Thread 1: crunching interpolation image line 20
Thread 2: crunching interpolation image line 25

... snip ...

Thread 2: crunching interpolation image line 1020
Preparing colormap.
Drawing interpolated power levels to map.
Floating point exception

Thorn · 03-01-2006

Eric, what's the hardware? Have you tried a recompile just for that hardware?

Floating point errors across different OSs indicate that you've got a bad math processor or one that isn't compatible with what has been compiled.

layer2 · 03-02-2006

Thorn,

I've tried this particular render on at least 6 different boxes, ranging from workstation grade P4's up to my high-end production servers. Fresh kismet compiles on every box, all with the same results.

The problem only occurs if I try running the interpolated power maps ( -p ) -- all other map types complete normally.

Are you interested in trying this on one of your boxes?

Thanks,
Eric

Dutch · 03-02-2006

Quote:

Originally Posted by layer2

Thorn,

I've tried this particular render on at least 6 different boxes, ranging from workstation grade P4's up to my high-end production servers. Fresh kismet compiles on every box, all with the same results.

The problem only occurs if I try running the interpolated power maps ( -p ) -- all other map types complete normally.

Are you interested in trying this on one of your boxes?

Thanks,
Eric

Couple of questions here..
Do the box you used in the above logsnippet, have more than one CPU ?
If you are running P4's are Hypertreading enabled ?
Which version of the GMP multiprecision lib has been used when compiling Kismet & gpsmap ?
Post the config.log from your kismet compilation.
And if you want, throw a link to a place where I can get the .gps and .xml files that you want tested with the interpolated power map.
It works fine here on a debian based box.

Dutch

layer2 · 03-02-2006

Dutch,

I'm using the GMP RPM from RHEL, which is currently at gmp-devel-4.1.2-5 on RHEL3 and gmp-devel-4.1.4-3 on RHEL4. Multiprocessor hyperthreaded boxes all around.

Here's the config.log from a RHEL3 box: http://www.presetkilllimit.us/config.log

Further testing indicates that the crash happens reproducibly for certain filter strings and not for others. When the render doesn't crash, the resulting map doesn't look like the samples I've seen elsewhere online. Here's the output of a run that didn't crash with the floating point error:

http://www.presetkilllimit.us/sample_map.png

Thanks,
Eric

Dutch · 03-02-2006

Quote:

Originally Posted by layer2

Dutch,

I'm using the GMP RPM from RHEL, which is currently at gmp-devel-4.1.2-5 on RHEL3 and gmp-devel-4.1.4-3 on RHEL4. Multiprocessor hyperthreaded boxes all around.

Here's the config.log from a RHEL3 box: http://www.presetkilllimit.us/config.log

Further testing indicates that the crash happens reproducibly for certain filter strings and not for others. When the render doesn't crash, the resulting map doesn't look like the samples I've seen elsewhere online. Here's the output of a run that didn't crash with the floating point error:

http://www.presetkilllimit.us/sample_map.png

Thanks,
Eric

Thats real weird, doesn't look like something should be wrong from the config.log.
Thinking that this might be a MP issue, have you tried specifying that only one thread should be used, or tried on a single processor box ?

The example you posted above, which command line parameters did you use ?
I'll try and post a gpsmap generated overlay with the same params, using the data and bitmap image I used in the Google Earth thread.

Dutch

layer2 · 03-02-2006

Dutch,

I've tried the command on single and multiprocessor boxes, with and without the -z switch, with the same results. The exact command I used to generate the sample png is:

gpsmap -D -z 4 -e -o sample.png -a -S 2 -G -b -u -U 20 -n 1 -L 0 -l ssid -f aa:bb:cc:dd:ee:ff -i -p *.gps

(I changed the MAC addr to protect the innocent)

Email me and we can make arrangements for you to get copies of the {gps,xml} files.

Thanks!

Dutch · 03-02-2006

Attached is an gpsmap example from the data used in the Visualizing Site Coverage with Google Earth thread.

I've used your commandline as stated above, with the addition of the track (-t & -Y 1), and using mapserver 6 (expedia) as no terraserver coverage of Denmark.

Dutch

03-01-2006	#1 (permalink)
layer2 PreSet Kill Limit Join Date: Mar 2006 Posts: 10	gpsmap: "Floating point exception" I've been unable to build any interpolated network power maps using gpsmap (from Kismet) version v2005.08.R1. ImageMagick is at 6.2.2. I've experienced the same floating point errors under Auditor, Fedora Core 4, RHEL3 and RHEL4. Any thoughts or advice appreciated. Thanks... $ gpsmap -D -z 4 -e -o out.png -f 00:0A:00:00:00:00 -i -p *.gps ... snip ... Converting map to greyscale. Calculating network coordinates and statistics... Processing 1 raw networks. Assigning network colors... Plotting 1 networks... Drawing network power interpolations... Interpolating power into graph points. Thread 0: crunching interpolation image line 0 Thread 1: crunching interpolation image line 5 Thread 2: crunching interpolation image line 10 Thread 3: crunching interpolation image line 15 Thread 1: crunching interpolation image line 20 Thread 2: crunching interpolation image line 25 ... snip ... Thread 2: crunching interpolation image line 1020 Preparing colormap. Drawing interpolated power levels to map. Floating point exception

03-01-2006	#2 (permalink)
Thorn Did you do the math? Join Date: Apr 2002 Location: Villa Straylight Posts: 10,097	Eric, what's the hardware? Have you tried a recompile just for that hardware? Floating point errors across different OSs indicate that you've got a bad math processor or one that isn't compatible with what has been compiled. __________________ Thorn "I'm The Doctor. I'm a Time Lord. I am from the planet Gallifrey in the constellation Kasterborous. I'm 903 years old and I am the man who is going to save your lives and all 6 billion people on the planet below... You got a problem with that?"

Thread Tools
Show Printable Version Email this Page
Display Modes
Linear Mode Switch to Hybrid Mode Switch to Threaded Mode

03-02-2006	#3 (permalink)
layer2 PreSet Kill Limit Join Date: Mar 2006 Posts: 10	Thorn, I've tried this particular render on at least 6 different boxes, ranging from workstation grade P4's up to my high-end production servers. Fresh kismet compiles on every box, all with the same results. The problem only occurs if I try running the interpolated power maps ( -p ) -- all other map types complete normally. Are you interested in trying this on one of your boxes? Thanks, Eric

03-02-2006	#5 (permalink)
layer2 PreSet Kill Limit Join Date: Mar 2006 Posts: 10	Dutch, I'm using the GMP RPM from RHEL, which is currently at gmp-devel-4.1.2-5 on RHEL3 and gmp-devel-4.1.4-3 on RHEL4. Multiprocessor hyperthreaded boxes all around. Here's the config.log from a RHEL3 box: http://www.presetkilllimit.us/config.log Further testing indicates that the crash happens reproducibly for certain filter strings and not for others. When the render doesn't crash, the resulting map doesn't look like the samples I've seen elsewhere online. Here's the output of a run that didn't crash with the floating point error: http://www.presetkilllimit.us/sample_map.png Thanks, Eric

03-02-2006	#7 (permalink)
layer2 PreSet Kill Limit Join Date: Mar 2006 Posts: 10	Dutch, I've tried the command on single and multiprocessor boxes, with and without the -z switch, with the same results. The exact command I used to generate the sample png is: gpsmap -D -z 4 -e -o sample.png -a -S 2 -G -b -u -U 20 -n 1 -L 0 -l ssid -f aa:bb:cc:dd:ee:ff -i -p *.gps (I changed the MAC addr to protect the innocent) Email me and we can make arrangements for you to get copies of the {gps,xml} files. Thanks!