Messages in this thread | | | Date | Fri, 7 Dec 2007 13:58:10 -0500 (EST) | From | Steven Rostedt <> | Subject | Re: Major regression on hackbench with SLUB |
| |
[ Note: I'm currently having problems with my email: http://www.domaindirect.com/network.html I'm in the process of setting up my own personal email server. ]
> > Can you do one run with oprofile, and see exactly where the cost is? It > > should hopefully be pretty darn obvious, considering your timing.
I kicked off my kernel tests which will take a few hours. I run "make -j256" (4*nr_CPUS) 10 times at SCHED_OTHER and 10 times at SCHED_FIFO (chrt -f 10) for both SLUB and SLAB.
This is automated, so I need to wait for it to finish before I can continue other tests.
> The biggest cost of __slab_alloc() in my profile is the "slab_lock()", but > that may not be the one that causes problems in a 64-cpu setup, so it > would be good to have that verified.
I'll run oprofile as soon as the kernel build test is done.
> case which can trigger on NUMA. That's another potential explanation of > why you'd see such a *huge* slowdown (ie if the whole node-match thing > doesn't work out, slub just never gets the fast-case at all). That said, > the number of places that actually pass a specific node to slub is very > limited, so I suspect it's not the node matching. But just disabling that > test in slab_alloc() might be one thing to test.
I'll try that after the oprofile.
Thanks,
-- Steve
| |