2002-09-23 05:39:06

by William Lee Irwin III

[permalink] [raw]
Subject: 2.5.38 vs. 2.5.38-mm1 dbench 512 oprofiles

2.5.38 virgin:
Throughput 34.0847 MB/sec (NB=42.6059 MB/sec 340.847 MBit/sec) 512 procs
c01053ec 332946298 89.2647 poll_idle
c0114a48 11069332 2.96775 load_balance
c015c196 6290235 1.68645 .text.lock.dcache
c0154783 4966046 1.33142 .text.lock.namei
c0114ec0 3330213 0.892848 scheduler_tick
c0131e94 2213627 0.593485 generic_file_write_nolock
c01a10cb 2120138 0.568421 .text.lock.dec_and_lock
c011734d 1064379 0.285366 .text.lock.sched
c0111728 777966 0.208577 smp_apic_timer_interrupt
c0131460 485556 0.13018 file_read_actor
c013f950 439352 0.117793 blk_queue_bounce
c0106418 417379 0.111902 .text.lock.semaphore
c010604c 327813 0.0878884 __down
c01145b8 312618 0.0838146 try_to_wake_up
c01a1050 282950 0.0758604 atomic_dec_and_lock
c01151e4 280517 0.0752081 do_schedule
c0151fb8 233523 0.0626088 path_lookup
c01513f4 204300 0.0547739 link_path_walk
c01449a0 197412 0.0529272 generic_file_llseek
c015b120 186943 0.0501204 d_alloc
c015b2e4 173003 0.046383 d_instantiate
c013fbf6 158793 0.0425733 .text.lock.highmem
c0130e08 150974 0.040477 find_get_page


2.5.38-mm1:
Throughput 20.966 MB/sec (NB=26.2075 MB/sec 209.66 MBit/sec) 512 procs
c01053ec 332946298 89.2647 poll_idle
c0114a48 11069332 2.96775 load_balance
c015c196 6290235 1.68645 .text.lock.dcache
c0154783 4966046 1.33142 .text.lock.namei
c0114ec0 3330213 0.892848 scheduler_tick
c0131e94 2213627 0.593485 generic_file_write_nolock
c01a10cb 2120138 0.568421 .text.lock.dec_and_lock
c011734d 1064379 0.285366 .text.lock.sched
c0111728 777966 0.208577 smp_apic_timer_interrupt
c0131460 485556 0.13018 file_read_actor
c013f950 439352 0.117793 blk_queue_bounce
c0106418 417379 0.111902 .text.lock.semaphore
c010604c 327813 0.0878884 __down
c01145b8 312618 0.0838146 try_to_wake_up
c01a1050 282950 0.0758604 atomic_dec_and_lock
c01151e4 280517 0.0752081 do_schedule
c0151fb8 233523 0.0626088 path_lookup
c01513f4 204300 0.0547739 link_path_walk
c01449a0 197412 0.0529272 generic_file_llseek
c015b120 186943 0.0501204 d_alloc
c015b2e4 173003 0.046383 d_instantiate
c013fbf6 158793 0.0425733 .text.lock.highmem
c0130e08 150974 0.040477 find_get_page


2002-09-23 05:49:10

by William Lee Irwin III

[permalink] [raw]
Subject: Re: 2.5.38 vs. 2.5.38-mm1 dbench 512 oprofiles

On Sun, Sep 22, 2002 at 10:36:00PM -0700, William Lee Irwin III wrote:
sorry real 2.5.38-mm1 profile:
c01053dc 884615995 93.5551 poll_idle
c0114c28 38573436 4.07944 load_balance
c01150a0 7901071 0.8356 scheduler_tick
c0111788 2659487 0.281262 smp_apic_timer_interrupt
c0131600 1442967 0.152605 file_read_actor
c01402a4 1199079 0.126812 blk_queue_bounce
c01320f4 1054365 0.111507 generic_file_write_nolock
c010d718 730020 0.0772053 timer_interrupt
c01a2abb 725817 0.0767608 .text.lock.dec_and_lock
c01175fb 674730 0.071358 .text.lock.sched
c0106408 356493 0.0377019 .text.lock.semaphore
c0115454 326529 0.034533 do_schedule
c011f6f0 309224 0.0327029 do_softirq
c0122eb4 262417 0.0277527 update_one_process
c0122fe8 222659 0.0235479 timer_bh
c010603c 191642 0.0202676 __down
c0114798 159658 0.0168851 try_to_wake_up
c015c66c 153232 0.0162055 d_lookup
c0145b50 141098 0.0149222 generic_file_llseek
c01391cc 123761 0.0130887 rmqueue
c01408d1 123358 0.0130461 .text.lock.highmem
c0147160 117950 0.0124741 get_empty_filp
c0174f00 105235 0.0111294 ext2_new_block

2002-09-23 09:00:13

by Dipankar Sarma

[permalink] [raw]
Subject: Re: 2.5.38 vs. 2.5.38-mm1 dbench 512 oprofiles

On Mon, Sep 23, 2002 at 06:02:15AM +0000, William Lee Irwin III wrote:
> On Sun, Sep 22, 2002 at 10:36:00PM -0700, William Lee Irwin III wrote:
> sorry real 2.5.38-mm1 profile:
> c01053dc 884615995 93.5551 poll_idle
> c0114c28 38573436 4.07944 load_balance

Any idea why scheduler time shoot up with mm1 ? The RCU core
patch (rcu_ltimer) in mm1 doesn't force any context switches, so it
can't be that.

Thanks
--
Dipankar Sarma <[email protected]> http://lse.sourceforge.net
Linux Technology Center, IBM Software Lab, Bangalore, India.

2002-09-23 13:54:20

by William Lee Irwin III

[permalink] [raw]
Subject: Re: 2.5.38 vs. 2.5.38-mm1 dbench 512 oprofiles

On Mon, Sep 23, 2002 at 06:02:15AM +0000, William Lee Irwin III wrote:
>> sorry real 2.5.38-mm1 profile:
>> c01053dc 884615995 93.5551 poll_idle
>> c0114c28 38573436 4.07944 load_balance

On Mon, Sep 23, 2002 at 02:40:55PM +0530, Dipankar Sarma wrote:
> Any idea why scheduler time shoot up with mm1 ? The RCU core
> patch (rcu_ltimer) in mm1 doesn't force any context switches, so it
> can't be that.

Basically, the time once spent in busywait becomes idle time.


Cheers,
Bill