I wrote:
> ...
> If you repeat this process several times:
>
> 1) Open Axiom
> 2) Allocate Matrix
> 3) Close Axiom
> 4) display residual and virtural memory
>
> For example do this 5 times in a row, do you find that the
> total available memory is constantly decreasing?
> ...
>
> Could you describe the test you use on GCL? What version of
> gcl are you using?

On Tuesday, June 21, 2005 9:10 PM Vanuxem Grégory wrote:

> (make-array '(1000 1000) :element-type 'long-float :initial-element 0)
>
> about 50 times
>

I GCL I am not certain that this function will necessarily allocate
all of the memory that might be required if each element of the
array was explicitly given some value.

> ...
> Sorry,
> 1) Open Axiom
> 4) display residual and virtural memory
> -- avoid printing
> new(1000,1000,0)$Matrix(SF);
> -- memory has increased of 10Mb
> new(1000,1000,0)$Matrix(SF);
> -- memory has increased of 10Mb
> new(1000,1000,0)$Matrix(SF);
> -- memory has increased of 10Mb
> new(1000,1000,0)$Matrix(SF);
> -- memory has increased of 10Mb
> new(1000,1000,0)$Matrix(SF);
> -- memory has increased of 10Mb
> new(1000,1000,0)$Matrix(SF);
> -- memory has increased of 10Mb
> new(1000,1000,0)$Matrix(SF);
> -- memory has increased of 10Mb
> .
> .
> .
> a lot of times (axiom will crash) => xmalloc failed

Why do say that this sort of test represents a "memory leak"?
If you do not close Axiom between these allocations of memory
you can not be testing for the conditional called a "memory leak".
A memory leak is a serious problem that occurs when a process does
not give back to the operating system some memory that it allocated
when it closes.

It seems quite normal to me that allocating many large arrays
eventually uses up all available virtual memory. It is possible
that Axiom is using a different data structure for representing
a Matrix then the GCL make-array.

Regards,
Bill Page.


_______________________________________________
Axiom-developer mailing list
Axiom-developer@nongnu.org
http://lists.nongnu.org/mailman/listinfo/axiom-developer

Reply via email to