On Aug 10, 2010, at 11:40 AM, objectwerks inc wrote:

> 
> On Aug 10, 2010, at 12:34 PM, Jonathon Kuo wrote:
> 
>> On Aug 10, 2010, at 11:13 AM, objectwerks inc wrote:
>> 
>>> (Why is this library thing important?  Each process that opens the library 
>>> has the space charged against the VM even though there is only one copy of 
>>> the library in memory with library code being shared across all processes 
>>> that use it)
>> 
>> Not disagreeing, but it's a bit hard to believe that Safari would need to 
>> map more and more libraries ad infinitum to the tune of 3GB+ worth of shared 
>> address space.
> 
> Did I ever say that?  I never said all that space is shared library memory 
> space.  I only said that some of it is and that the actual VSIZE numbers are 
> irrelevant to the discussion, among which reasons was the shared library 
> space being counted.
> 
>> And even harder to understand why it can never relinquish any of it, even 
>> when all windows and tabs are closed. Keep Safari up long enough, it becomes 
>> slower and slower as the VMSIZE value increases until it exceeds the process 
>> VM limit. When Safari is running this fat, manually closing a window can 
>> take minutes (and pegs the CPU while doing so).
>> 
> 
> yes, this does sound like a design bug.  I never claimed that there were not 
> problems.  I have noticed them myself and not just with 5.x Safari (and in  
> fact have not noticed them yet with 5.x).
> 
> I am wondering if when you close all your tabs and windows, does it NEVER 
> reduce its footprint?  I am wondering if there is a delay in releasing stuff 
> as it may be more efficient to reuse already allocated space.  I am betting 
> no one here has done any rigorous testing.  I know I have not.

I haven't done rigorous testing, but the times where I have closed all Safari 
windows and cleared the cache, it didn't reduce the reported VSIZE except by 
.01GB or so, and then just sat there.
> 
> My point is that looking at VSIZE itself tells you nothing and your VSIZE and 
> my VSIZE may be totally different from one another when the problems start.  
> Let's not get hung up on VSIZE as it is an irrelevant number for this 
> discussion.

No process can exceed its per-process VM limit, so the closer the reported 
VSIZE approaches 4GB, the closer is its demise.

_______________________________________________
MacOSX-talk mailing list
[email protected]
http://www.omnigroup.com/mailman/listinfo/macosx-talk

Reply via email to