|No such thing as a small change|
Re^6: Timing of garbage collectionby BrowserUk (Pope)
|on Jan 19, 2013 at 17:21 UTC||Need Help??|
First, let me say that I was not arguing with the basic premise that in general, Perl does not give memory back to the OS. Only that the definitive statement (in common with most definitive statements), that perl *never* gives memory back to the OS, is not strictly true.
I think that it is fair to say that "mileage varies" even in this simple test....
Hm. If by that you mean that the exact numbers of the recorded memory allocation can vary a little. Yes, of course.
You're running the tests on a system (OS; file-cache; virtual memory; background processes and device drivers et al.) -- that is ever in flux, but that does not detract from the fact that (on AS Perl on windows), if you make single allocations of greater than 1 MB, when those allocations are freed, they will be returned to the OS. They will never be fragmented, because they will never be reallocated for anything else.
This doesn't just apply to large scalars either. For example, if you create an array where the base AV requires more than 1 MB of contiguous memory, when that array is freed, that contiguous chunk for the AV gets returned to the OS.
Here I preallocate the AV for an array to hold upto 1 million values which on my 64-bit perl requires an 8*1e6 = 8000000 byte chunk of contiguous memory:
And when the array is freed, that ~8MB is given back.
And if I populate that array:
Although the additional 24MB of ram (1e6 * 24B) required to hold the array's scalars is not returned to the OS, the 8MB allocated to the AV, *still is*.
The difference is that the 1e6 x 24 byte NVs are allocated from the process memory pool on the fly, and that pool is expanded -- in smallish chunks -- on the fly to accommodate them. Thus those piecemeal allocations are not contiguous chunks greater than 1MB, so do not get returned to the OS.
This knowledge is arcane and system dependent, but with care (and the need) it can be exploited to good effect if you use a system where it is so.
Just beware of overly simplistic, 'definitive truths'.
With the rise and rise of 'Social' network sites: 'Computers are making people easier to use everyday'
Examine what is said, not who speaks -- Silence betokens consent -- Love the truth but pardon error.
"Science is about questioning the status quo. Questioning authority".
In the absence of evidence, opinion is indistinguishable from prejudice.