I think the vma sorting is hurting (contention on i_shared_sem). It also
only gains a factor of about 2 in the corner case you were worried about
... any ideas on ways to make it cheaper? We need something more dramatic
anyway (though the other solutions I have revolve around the same
contention problem ;-( ) I just backed it out for now in my stuff.
Note systimes on kernel compiles, and SDET throughput.
M.
Kernbench: (make -j N vmlinux, where N = 2 x num_cpus)
Elapsed System User CPU
2.5.67-mm1 44.02 80.92 567.78 1473.00
2.5.68-mm1 44.14 82.82 567.21 1472.25
Kernbench: (make -j N vmlinux, where N = 16 x num_cpus)
Elapsed System User CPU
2.5.67-mm1 44.76 89.40 576.67 1487.50
2.5.68-mm1 44.84 95.52 576.48 1498.25
Kernbench: (make -j vmlinux, maximal tasks)
Elapsed System User CPU
2.5.67-mm1 44.77 87.34 576.85 1482.75
2.5.68-mm1 45.18 96.24 576.75 1489.75
DISCLAIMER: SPEC(tm) and the benchmark name SDET(tm) are registered
trademarks of the Standard Performance Evaluation Corporation. This
benchmarking was performed for research purposes only, and the run results
are non-compliant and not-comparable with any published results.
Results are shown as percentages of the first set displayed
SDET 1 (see disclaimer)
Throughput Std. Dev
2.5.67-mm1 100.0% 2.2%
2.5.68-mm1 103.5% 2.3%
SDET 2 (see disclaimer)
Throughput Std. Dev
2.5.67-mm1 100.0% 3.9%
2.5.68-mm1 101.8% 2.7%
SDET 4 (see disclaimer)
Throughput Std. Dev
2.5.67-mm1 100.0% 0.9%
2.5.68-mm1 98.7% 2.3%
SDET 8 (see disclaimer)
Throughput Std. Dev
2.5.67-mm1 100.0% 0.9%
2.5.68-mm1 96.9% 0.9%
SDET 16 (see disclaimer)
Throughput Std. Dev
2.5.67-mm1 100.0% 0.8%
2.5.68-mm1 89.7% 0.2%
SDET 32 (see disclaimer)
Throughput Std. Dev
2.5.67-mm1 100.0% 0.3%
2.5.68-mm1 76.6% 0.4%
SDET 64 (see disclaimer)
Throughput Std. Dev
2.5.67-mm1 100.0% 0.3%
2.5.68-mm1 75.5% 0.1%
SDET 128 (see disclaimer)
Throughput Std. Dev
2.5.67-mm1 100.0% 0.2%
2.5.68-mm1 78.7% 0.2%
diffprofile 2.5.6{7,8}-mm1/kernbench/0/profile
6159 4.4% total
3786 0.0% move_vma_start
3158 3222.4% __vma_link
762 8466.7% __down
522 293.3% may_open
478 16.5% zap_pte_range
458 74.2% __wake_up
438 13.2% find_get_page
400 102.3% remove_shared_vm_struct
332 23.1% schedule
282 0.0% install_page
257 4.9% __copy_to_user_ll
255 7.7% __copy_from_user_ll
198 47.9% copy_page_range
133 21.6% sys_brk
115 5.2% page_remove_rmap
...
-100 -0.7% do_anonymous_page
-100 -8.3% file_move
-127 -10.7% link_path_walk
-128 -6.5% path_lookup
-179 -2.2% __d_lookup
-224 -72.3% do_lookup
-433 -21.4% do_page_fault
-447 -57.7% filemap_nopage
-493 -25.1% do_no_page
-535 -85.3% dentry_open
-776 -38.5% page_add_rmap
-2452 -4.8% default_idle
diffprofile 2.5.6{7,8}-mm1/sdetbench/128/profile
117130 18.0% total
92204 26.2% default_idle
28537 962.5% __down
9253 378.1% schedule
9119 596.4% __wake_up
6573 0.0% move_vma_start
4587 1176.2% __vma_link
1899 45.7% remove_shared_vm_struct
1805 47.2% do_wp_page
1467 11.0% copy_page_range
978 30.6% copy_mm
506 68.2% vma_link
490 0.0% install_page
395 411.5% rb_insert_color
354 2.0% zap_pte_range
347 55.3% task_mem
320 29.4% __block_prepare_write
314 52.7% may_open
313 2235.7% cpu_idle
304 33.6% __copy_user_intel
294 490.0% default_wake_function
276 8.2% do_anonymous_page
219 509.3% __vma_link_rb
...
-206 -17.5% ext2_new_inode
-208 -14.1% number
-208 -100.0% pid_revalidate
-209 -22.9% update_atime
-214 -41.9% d_path
-222 -26.7% proc_root_link
-231 -20.0% __mark_inode_dirty
-234 -13.9% release_task
-238 -10.5% do_page_fault
-244 -47.7% ext2_new_block
-253 -1.8% page_remove_rmap
-259 -62.3% do_lookup
-271 -22.7% file_kill
-287 -39.9% ext2_get_group_desc
-289 -15.7% __find_get_block
-292 -43.4% __find_get_block_slow
-292 -30.4% dput
-296 -21.0% __brelse
-312 -36.4% d_instantiate
-313 -12.6% link_path_walk
-348 -39.1% prune_dcache
-352 -18.6% current_kernel_time
-358 -24.4% fd_install
-369 -25.7% __fput
-410 -34.9% flush_signal_handlers
-424 -28.0% .text.lock.highmem
-450 -65.5% dentry_open
-456 -24.4% d_alloc
-475 -45.0% proc_check_root
-525 -21.6% file_move
-555 -21.0% proc_pid_stat
-631 -38.6% grab_block
-643 -49.5% .text.lock.attr
-648 -7.6% release_pages
-675 -45.0% .text.lock.file_table
-718 -47.7% .text.lock.base
-749 -27.9% kmem_cache_free
-788 -48.2% __read_lock_failed
-882 -38.5% filemap_nopage
-918 -35.3% path_release
-947 -20.8% .text.lock.filemap
-994 -22.3% do_no_page
-1037 -93.8% task_vsize
-1068 -13.1% page_add_rmap
-1591 -13.8% __d_lookup
-1830 -27.6% path_lookup
-2116 -61.3% .text.lock.namei
-2399 -39.5% follow_mount
-2987 -61.0% .text.lock.dcache
-3705 -38.0% atomic_dec_and_lock
-4444 -61.7% .text.lock.dec_and_lock
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Wed Apr 23 2003 - 22:00:29 EST