lmbench context switch regression
Hi,
I'm seeing quite a large context switch speed regression as reported by lmbench when I patched from test9 to test9-mm2.
The obvious thing I can see from the patch is the PF_DEAD finish_task_switch change. I don't have time to investigate further tonight though.
lmbench two 0 sized processes context switch times go from around 1.80us to 2.60us on my PIII 650 (UP).
Quoted 9 lines · Nick Piggin
Hi,
I'm seeing quite a large context switch speed regression as reported by lmbench when I patched from test9 to test9-mm2.The obvious thing I can see from the patch is the
PF_DEADfinish_task_switchchange. I don't have time to investigate further tonight though.lmbench two 0 sized processes context switch times go from around 1.80us to 2.60us on my PIII 650 (UP).
The regression is smaller when going from test9 to test9-bk. About .2us (10%). It does not go away when reverting the PF_DEAD change. Maybe its just lucky alignment?