Segmentation fault

[PhilF]

I have an overclocked Pentium 4 computer that is exhibiting a curious behavior.

This machine runs mprime 24/7. It also runs a shell program that uses Samba (smbclient specifically) to gather data from other Linux nodes to produce a report.

On warm days, when the CPU temperature is above 60C degrees (but well below the maximum allowable), smbclient will occasionally give a non-fatal segmentation fault. On cooler days when the CPU temperature is below 60C, it never does that.

I am wondering, from a hardware perspective, what could be going on that would cause these segmentation faults from smbclient, yet mprime never gives any errors, and in fact always returns good results when assigned double checks?

Any ideas?

[xilman]

Originally Posted by PhilF

I have an overclocked Pentium 4 computer that is exhibiting a curious behavior.

This machine runs mprime 24/7. It also runs a shell program that uses Samba (smbclient specifically) to gather data from other Linux nodes to produce a report.

On warm days, when the CPU temperature is above 60C degrees (but well below the maximum allowable), smbclient will occasionally give a non-fatal segmentation fault. On cooler days when the CPU temperature is below 60C, it never does that.

I am wondering, from a hardware perspective, what could be going on that would cause these segmentation faults from smbclient, yet mprime never gives any errors, and in fact always returns good results when assigned double checks?

Any ideas?

Accessing the disk, or a different disk? Using a different chunk of memory?

Paul

[PhilF]

I don't think it is disk access related, since it never happens when the processor is cool.

But you might be onto something as far as it being related to a different chunk of memory. It is another warm day today, and the segmentation faults returned. I then stopped mprime and started a torture test. It started with a 1024K FFT size, and the segmentation faults went away. I just restarted the torture test with a MinTortureFFT of 1792K, and sure enough, the segmentation faults returned.

I am still a little bewildered that a program can have a heat related problem when mprime is not having any problems at all. I just always assumed mprime is going to be more sensitive to any sort of processor/memory issues than any other program.

I just had a thought about priority. Since smbclient runs at a higher priority than mprime, maybe somehow that is why it is the program that always causes and/or catches the segmentation faults.

[xilman]

Originally Posted by PhilF

I don't think it is disk access related, since it never happens when the processor is cool.

But you might be onto something as far as it being related to a different chunk of memory. It is another warm day today, and the segmentation faults returned. I then stopped mprime and started a torture test. It started with a 1024K FFT size, and the segmentation faults went away. I just restarted the torture test with a MinTortureFFT of 1792K, and sure enough, the segmentation faults returned.

I am still a little bewildered that a program can have a heat related problem when mprime is not having any problems at all. I just always assumed mprime is going to be more sensitive to any sort of processor/memory issues than any other program.

I just had a thought about priority. Since smbclient runs at a higher priority than mprime, maybe somehow that is why it is the program that always causes and/or catches the segmentation faults.

If it is the memory itself that is heat sensitive, and not the cpu, then only the process which actually uses that memory will trigger the fault.


Paul

[patrik]

Originally Posted by PhilF

I am still a little bewildered that a program can have a heat related problem when mprime is not having any problems at all. I just always assumed mprime is going to be more sensitive to any sort of processor/memory issues than any other program.

I had a similar problem with GMP-ECM causing segfaults while at the same time mprime was running without problems. I had to lower my overclock before I got rid of them. Don't remember if it was correlated with heat.

(Btw., is Oklahoma so warm in the winter?)

[PhilF]

Originally Posted by patrik

(Btw., is Oklahoma so warm in the winter?)

It is not supposed to be, but this year we are setting record high temperatures, and wildfires are rampant.