[cmucl-imp] Working set

[Raymond Toy]

On 11/6/10 8:29 AM, Helmut Eller wrote:
> * Raymond Toy [2010-11-06 11:58] writes:
>
>> On 11/5/10 3:41 PM, Helmut Eller wrote:
>>> Is there an easy way to determine the working set size (resident size)
>>> that is needed to evaluate some form?  
>>>
>>> TIME prints various things but not what I need.  The Linux top command
>>> shows a column with the resident size; essentially I'd like to see
>>> the maximum of the those values sampled while executing the form.
>>>
>>> Does the GC have this information already?  Would it be
>>> difficult/expensive to use getrusage in GC hooks to collect the
>>> information?
>> I don't see why you can't use getrusage in a GC hook to get his
>> information.  After all, you made slime use a GC hook to show how much
>> memory is used in each generation when a GC happens.  Would summing up
>> the memory usage in each generation be close enough to the resident size
>> that you want?
> Would that include the space needed to run the GC?  Doesn't the GC need
> about twice us much while moving from-space to to-space?
No, I don't think the space needed to run the GC is included.  Since
it's a copying GC, I think the extra space needed is equal to the space
retained; the stuff that is not copied is garbage.

The *gc-notify-after* hook is given the bytes retained and the bytes
freed.  I guess the bytes retained would be the extra space needed for
GCing.   I don't think this includes the space needed by the static
space, or the read-only space.

Note that there is a space profiler too that Martin contributed many
years ago.  (See profile::with-space-reports.)  I've never actually used
this before.

Ray