Re: profiling the kernel
Re: profiling the kernel
- Subject: Re: profiling the kernel
- From: chuck remes <email@hidden>
- Date: Mon, 23 Feb 2004 17:01:43 -0600
On Feb 23, 2004, at 4:54 PM, chuck remes wrote:
While doing some performance tuning on a driver, I noticed that
darwin-x86 is significantly slower than darwin-ppc on roughly
equivalent hardware. When doing some pretty common operations (like
running "sum <600 MB file>") most of the system time is charged to the
kernel (e.g. kernel_task).
If I wanted to profile the kernel and try to narrow down the sections
of code that are problematic, how would I go about doing this? Is it
safe to assume that 2-machine debugging is a necessary prerequisite?
Is there any chance that the tools Apple uses internally for this
operation will see the light of day?
If this is NOT feasible, can any Apple kernel engineers suggest places
in the code that could use some scrutiny?
Any and all suggestions are welcome.
Five minutes after sending this (ain't that always the way it goes?) I
stumbled across "kgmon" which looks like it does what I want. Is this
the right place to start?
If so, all I need is a kernel and its support frameworks compiled with
debugging symbols on x86... (I'll get help from Felix@opendarwin with
this part).
cr
_______________________________________________
darwin-kernel mailing list | email@hidden
Help/Unsubscribe/Archives:
http://www.lists.apple.com/mailman/listinfo/darwin-kernel
Do not post admin requests to the list. They will be ignored.