* [0/3] filtered wakeups respun @ 2004-05-05 6:06 William Lee Irwin III 2004-05-05 6:08 ` William Lee Irwin III 0 siblings, 1 reply; 6+ messages in thread From: William Lee Irwin III @ 2004-05-05 6:06 UTC (permalink / raw) To: akpm; +Cc: linux-kernel [1/3]: filtered wakeups filter wakeups by the page being woken up for [2/3]: filtered buffers filter wakeups by the bh being woken up for [3/3]: wakeone restore wake-one semantics to bitlocking for pages and bh's Same machine/etc. as before, except this time, ext3 instead of ext2. ext3 shows noise-level differences in raw throughputs with large reductions in cpu overhead, mostly on the read side. ext2 results differ from these in that a 23% boost to sequential write cpu efficiency (throughput scaled by %cpu) is also achieved for sequential writes, almost entirely due to wake-one semantics. The tests take long enough to run that I've not done the ext2 results on a precisely-matching codebase. From the extant ext2 results: $ cat ~/tmp/virgin_mm.log/tiotest.log Tiotest results for 512 concurrent io threads: ,----------------------------------------------------------------------. | Item | Time | Rate | Usr CPU | Sys CPU | +-----------------------+----------+--------------+----------+---------+ | Write 16384 MBs | 1118.1 s | 14.654 MB/s | 1.6 % | 280.9 % | | Random Write 2000 MBs | 336.2 s | 5.950 MB/s | 0.8 % | 20.4 % | | Read 16384 MBs | 1717.1 s | 9.542 MB/s | 1.4 % | 31.8 % | | Random Read 2000 MBs | 465.2 s | 4.300 MB/s | 1.1 % | 36.1 % | `----------------------------------------------------------------------' $ cat ~/tmp/filtered_wakeup.log/tiotest.log Tiotest results for 512 concurrent io threads: ,----------------------------------------------------------------------. | Item | Time | Rate | Usr CPU | Sys CPU | +-----------------------+----------+--------------+----------+---------+ | Write 16384 MBs | 1099.5 s | 14.901 MB/s | 2.2 % | 279.3 % | | Random Write 2000 MBs | 333.8 s | 5.991 MB/s | 1.0 % | 14.9 % | | Read 16384 MBs | 1706.3 s | 9.602 MB/s | 1.4 % | 19.1 % | | Random Read 2000 MBs | 460.3 s | 4.345 MB/s | 1.1 % | 14.8 % | `----------------------------------------------------------------------' $ cat ~/tmp/wakeone.log/tiotest.log Tiotest results for 512 concurrent io threads: ,----------------------------------------------------------------------. | Item | Time | Rate | Usr CPU | Sys CPU | +-----------------------+----------+--------------+----------+---------+ | Write 16384 MBs | 1073.8 s | 15.258 MB/s | 1.5 % | 237.3 % | | Random Write 2000 MBs | 336.9 s | 5.937 MB/s | 0.9 % | 15.2 % | | Read 16384 MBs | 1703.0 s | 9.621 MB/s | 1.3 % | 18.8 % | | Random Read 2000 MBs | 458.6 s | 4.361 MB/s | 1.0 % | 14.9 % | `----------------------------------------------------------------------' /home/wli/tmp/virgin_mm.log/tiotest.log: Write: 5.1873MB/cpusec Random Write: 28.0660MB/cpusec Read: 28.7410MB/cpusec Random Read: 11.5591MB/cpusec /home/wli/tmp/filtered_wakeup.log/tiotest.log: Write: 5.2934MB/cpusec Random Write: 37.6792MB/cpusec Read: 46.8390MB/cpusec Random Read: 27.3270MB/cpusec /home/wli/tmp/wakeone.log/tiotest.log: Write: 6.3894MB/cpusec Random Write: 36.8758MB/cpusec Read: 47.8657MB/cpusec Random Read: 27.4277MB/cpusec The wakeone implementation used for the ext2 run(s) above was somewhat less refined than the current one in that it didn't implement wake-one semantics for lock_buffer() and committed a major stupidity in waking more waiters than necessary in its wake_up_filtered(). One should also note specific complaints about random read performance are going around, and this near triples ext3's cpu efficiency on random reads i.e. it takes ext3 from 10.3MB/cpusec to 28.5MB/cpusec. ext3 results; before: Tiotest results for 512 concurrent io threads: ,----------------------------------------------------------------------. | Item | Time | Rate | Usr CPU | Sys CPU | +-----------------------+----------+--------------+----------+---------+ | Write 16384 MBs | 926.5 s | 17.683 MB/s | 1.9 % | 161.3 % | | Random Write 2000 MBs | 333.5 s | 5.998 MB/s | 0.9 % | 21.0 % | | Read 16384 MBs | 1634.0 s | 10.027 MB/s | 1.5 % | 28.4 % | | Random Read 2000 MBs | 448.1 s | 4.463 MB/s | 1.2 % | 42.2 % | `----------------------------------------------------------------------' Throughput scaled by cpu consumption: Write: 10.8352MB/cpusec Random Write: 27.3881MB/cpusec Read: 33.5351MB/cpusec Random Read: 10.2834MB/cpusec top 10 cpu consumers: 15328 finish_task_switch 79.8333 10149 __wake_up 158.5781 9859 generic_file_aio_write_nolock 4.3393 8836 file_read_actor 39.4464 7601 __do_softirq 26.3924 3114 kmem_cache_free 24.3281 2810 __find_get_block 9.7569 2727 prepare_to_wait 21.3047 2464 kmem_cache_alloc 19.2500 1675 tl0_linux32 52.3438 top 10 scheduler callers: 8827430 wait_on_page_bit 30650.7986 327735 __wait_on_buffer 1463.1027 209926 __handle_preemption 13120.3750 138613 worker_thread 254.8033 35838 generic_file_aio_write_nolock 15.7738 32265 __lock_page 112.0312 16281 pipe_wait 127.1953 9538 do_exit 9.3145 7622 shrink_list 4.1067 6816 compat_sys_nanosleep 17.7500 after: Tiotest results for 512 concurrent io threads: ,----------------------------------------------------------------------. | Item | Time | Rate | Usr CPU | Sys CPU | +-----------------------+----------+--------------+----------+---------+ | Write 16384 MBs | 926.7 s | 17.680 MB/s | 1.9 % | 140.4 % | | Random Write 2000 MBs | 334.4 s | 5.981 MB/s | 0.9 % | 19.7 % | | Read 16384 MBs | 1649.8 s | 9.931 MB/s | 1.3 % | 19.0 % | | Random Read 2000 MBs | 443.6 s | 4.509 MB/s | 1.1 % | 14.7 % | `----------------------------------------------------------------------' Throughput scaled by cpu consumption: Write: 12.4245MB/cpusec Random Write: 29.0340MB/cpusec Read: 48.9212MB/cpusec Random Read: 28.5380MB/cpusec top 10 cpu consumers: 9751 generic_file_aio_write_nolock 4.2918 9116 file_read_actor 40.6964 7419 __do_softirq 25.7604 5217 finish_task_switch 27.1719 3482 __find_get_block 12.0903 2725 kmem_cache_free 21.2891 2669 wake_up_filtered 13.9010 2543 kmem_cache_alloc 19.8672 1629 find_get_page 16.9688 1613 tl0_linux32 50.4062 top 10 scheduler callers: 2402700 wait_on_page_bit 6825.8523 198357 __handle_preemption 12397.3125 179318 worker_thread 329.6287 18343 generic_file_aio_write_nolock 8.0735 15687 pipe_wait 122.5547 9306 do_exit 9.0879 7531 __lock_buffer 39.2240 6814 compat_sys_nanosleep 17.7448 6716 kswapd 29.9821 5429 sys_wait4 9.4253 -- wli ^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [0/3] filtered wakeups respun 2004-05-05 6:06 [0/3] filtered wakeups respun William Lee Irwin III @ 2004-05-05 6:08 ` William Lee Irwin III 2004-05-05 6:11 ` [2/3] filtered buffer_head wakeups William Lee Irwin III 0 siblings, 1 reply; 6+ messages in thread From: William Lee Irwin III @ 2004-05-05 6:08 UTC (permalink / raw) To: akpm, linux-kernel On Tue, May 04, 2004 at 11:06:12PM -0700, William Lee Irwin III wrote: > Same machine/etc. as before, except this time, ext3 instead of ext2. > ext3 shows noise-level differences in raw throughputs with large > reductions in cpu overhead, mostly on the read side. Precisely the same filtered wakeups for pages as before. Drop in a fresh wakeup primitive that uses a key to discriminate between the waiters for different objects on a hashed waitqueue, and make the page waiting functions use it. -- wli Index: wake-2.6.6-rc3-mm1/include/linux/wait.h =================================================================== --- wake-2.6.6-rc3-mm1.orig/include/linux/wait.h 2004-04-03 19:37:07.000000000 -0800 +++ wake-2.6.6-rc3-mm1/include/linux/wait.h 2004-05-04 13:16:00.000000000 -0700 @@ -28,6 +28,11 @@ struct list_head task_list; }; +struct filtered_wait_queue { + void *key; + wait_queue_t wait; +}; + struct __wait_queue_head { spinlock_t lock; struct list_head task_list; @@ -104,6 +109,7 @@ list_del(&old->task_list); } +void FASTCALL(wake_up_filtered(wait_queue_head_t *, void *)); extern void FASTCALL(__wake_up(wait_queue_head_t *q, unsigned int mode, int nr)); extern void FASTCALL(__wake_up_locked(wait_queue_head_t *q, unsigned int mode)); extern void FASTCALL(__wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr)); @@ -257,6 +263,16 @@ wait->func = autoremove_wake_function; \ INIT_LIST_HEAD(&wait->task_list); \ } while (0) + +#define DEFINE_FILTERED_WAIT(name, p) \ + struct filtered_wait_queue name = { \ + .key = p, \ + .wait = { \ + .task = current, \ + .func = autoremove_wake_function, \ + .task_list = LIST_HEAD_INIT(name.wait.task_list),\ + }, \ + } #endif /* __KERNEL__ */ Index: wake-2.6.6-rc3-mm1/kernel/sched.c =================================================================== --- wake-2.6.6-rc3-mm1.orig/kernel/sched.c 2004-04-30 15:06:49.000000000 -0700 +++ wake-2.6.6-rc3-mm1/kernel/sched.c 2004-05-04 13:16:00.000000000 -0700 @@ -2518,6 +2518,19 @@ } } +void fastcall wake_up_filtered(wait_queue_head_t *q, void *key) +{ + unsigned long flags; + unsigned int mode = TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE; + struct filtered_wait_queue *wait, *save; + spin_lock_irqsave(&q->lock, flags); + list_for_each_entry_safe(wait, save, &q->task_list, wait.task_list) { + if (wait->key == key) + wait->wait.func(&wait->wait, mode, 0); + } + spin_unlock_irqrestore(&q->lock, flags); +} + /** * __wake_up - wake up threads blocked on a waitqueue. * @q: the waitqueue Index: wake-2.6.6-rc3-mm1/mm/filemap.c =================================================================== --- wake-2.6.6-rc3-mm1.orig/mm/filemap.c 2004-04-30 15:06:49.000000000 -0700 +++ wake-2.6.6-rc3-mm1/mm/filemap.c 2004-05-04 13:16:00.000000000 -0700 @@ -307,16 +307,16 @@ void fastcall wait_on_page_bit(struct page *page, int bit_nr) { wait_queue_head_t *waitqueue = page_waitqueue(page); - DEFINE_WAIT(wait); + DEFINE_FILTERED_WAIT(wait, page); do { - prepare_to_wait(waitqueue, &wait, TASK_UNINTERRUPTIBLE); + prepare_to_wait(waitqueue, &wait.wait, TASK_UNINTERRUPTIBLE); if (test_bit(bit_nr, &page->flags)) { sync_page(page); io_schedule(); } } while (test_bit(bit_nr, &page->flags)); - finish_wait(waitqueue, &wait); + finish_wait(waitqueue, &wait.wait); } EXPORT_SYMBOL(wait_on_page_bit); @@ -344,7 +344,7 @@ BUG(); smp_mb__after_clear_bit(); if (waitqueue_active(waitqueue)) - wake_up_all(waitqueue); + wake_up_filtered(waitqueue, page); } EXPORT_SYMBOL(unlock_page); @@ -363,7 +363,7 @@ smp_mb__after_clear_bit(); } if (waitqueue_active(waitqueue)) - wake_up_all(waitqueue); + wake_up_filtered(waitqueue, page); } EXPORT_SYMBOL(end_page_writeback); @@ -379,16 +379,16 @@ void fastcall __lock_page(struct page *page) { wait_queue_head_t *wqh = page_waitqueue(page); - DEFINE_WAIT(wait); + DEFINE_FILTERED_WAIT(wait, page); while (TestSetPageLocked(page)) { - prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); + prepare_to_wait(wqh, &wait.wait, TASK_UNINTERRUPTIBLE); if (PageLocked(page)) { sync_page(page); io_schedule(); } } - finish_wait(wqh, &wait); + finish_wait(wqh, &wait.wait); } EXPORT_SYMBOL(__lock_page); ^ permalink raw reply [flat|nested] 6+ messages in thread
* [2/3] filtered buffer_head wakeups 2004-05-05 6:08 ` William Lee Irwin III @ 2004-05-05 6:11 ` William Lee Irwin III 2004-05-05 6:16 ` [3/3] wake-one PG_locked/BH_Lock semantics William Lee Irwin III 0 siblings, 1 reply; 6+ messages in thread From: William Lee Irwin III @ 2004-05-05 6:11 UTC (permalink / raw) To: akpm, linux-kernel On Tue, May 04, 2004 at 11:08:49PM -0700, William Lee Irwin III wrote: > Precisely the same filtered wakeups for pages as before. Drop in a > fresh wakeup primitive that uses a key to discriminate between the > waiters for different objects on a hashed waitqueue, and make the > page waiting functions use it. Now, make bh's use the new wakeup primitive also. This has the bugfix vs. the prior version that autoremoved waitqueue wakeup functions are made to match autoremove API usage in __wait_event_filtered(). -- wli Index: wake-2.6.6-rc3-mm1/fs/buffer.c =================================================================== --- wake-2.6.6-rc3-mm1.orig/fs/buffer.c 2004-04-30 15:06:46.000000000 -0700 +++ wake-2.6.6-rc3-mm1/fs/buffer.c 2004-05-04 13:16:16.000000000 -0700 @@ -74,7 +74,7 @@ smp_mb(); if (waitqueue_active(wq)) - wake_up_all(wq); + wake_up_filtered(wq, bh); } EXPORT_SYMBOL(wake_up_buffer); @@ -93,10 +93,10 @@ void __wait_on_buffer(struct buffer_head * bh) { wait_queue_head_t *wqh = bh_waitq_head(bh); - DEFINE_WAIT(wait); + DEFINE_FILTERED_WAIT(wait, bh); do { - prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); + prepare_to_wait(wqh, &wait.wait, TASK_UNINTERRUPTIBLE); if (buffer_locked(bh)) { struct block_device *bd; smp_mb(); @@ -106,7 +106,7 @@ io_schedule(); } } while (buffer_locked(bh)); - finish_wait(wqh, &wait); + finish_wait(wqh, &wait.wait); } static void Index: wake-2.6.6-rc3-mm1/fs/jbd/transaction.c =================================================================== --- wake-2.6.6-rc3-mm1.orig/fs/jbd/transaction.c 2004-04-30 15:06:46.000000000 -0700 +++ wake-2.6.6-rc3-mm1/fs/jbd/transaction.c 2004-05-04 13:16:16.000000000 -0700 @@ -638,7 +638,7 @@ jbd_unlock_bh_state(bh); /* commit wakes up all shadow buffers after IO */ wqh = bh_waitq_head(jh2bh(jh)); - wait_event(*wqh, (jh->b_jlist != BJ_Shadow)); + wait_event_filtered(*wqh, jh2bh(jh), (jh->b_jlist != BJ_Shadow)); goto repeat; } Index: wake-2.6.6-rc3-mm1/include/linux/wait.h =================================================================== --- wake-2.6.6-rc3-mm1.orig/include/linux/wait.h 2004-05-04 13:16:00.000000000 -0700 +++ wake-2.6.6-rc3-mm1/include/linux/wait.h 2004-05-04 13:16:27.000000000 -0700 @@ -146,7 +146,6 @@ break; \ __wait_event(wq, condition); \ } while (0) - #define __wait_event_interruptible(wq, condition, ret) \ do { \ wait_queue_t __wait; \ @@ -273,7 +272,28 @@ .task_list = LIST_HEAD_INIT(name.wait.task_list),\ }, \ } - + +#define __wait_event_filtered(wq, key, condition) \ +do { \ + DEFINE_FILTERED_WAIT(__wait, key); \ + wait_queue_head_t *__wqh = &(wq); \ + wait_queue_t *__wqe = &__wait.wait; \ + for (;;) { \ + prepare_to_wait(__wqh, __wqe, TASK_UNINTERRUPTIBLE); \ + if (condition) \ + break; \ + schedule(); \ + } \ + finish_wait(__wqh, __wqe); \ +} while (0) + + +#define wait_event_filtered(wq, key, condition) \ +do { \ + if (!(condition)) \ + __wait_event_filtered(wq, key, condition); \ +} while (0) + #endif /* __KERNEL__ */ #endif ^ permalink raw reply [flat|nested] 6+ messages in thread
* [3/3] wake-one PG_locked/BH_Lock semantics 2004-05-05 6:11 ` [2/3] filtered buffer_head wakeups William Lee Irwin III @ 2004-05-05 6:16 ` William Lee Irwin III 2004-05-05 6:42 ` Michael J. Cohen 0 siblings, 1 reply; 6+ messages in thread From: William Lee Irwin III @ 2004-05-05 6:16 UTC (permalink / raw) To: akpm, linux-kernel On Tue, May 04, 2004 at 11:11:21PM -0700, William Lee Irwin III wrote: > Now, make bh's use the new wakeup primitive also. This has the bugfix > vs. the prior version that autoremoved waitqueue wakeup functions are > made to match autoremove API usage in __wait_event_filtered(). This is still grossly inefficient in that it's only necessary to wake one waiter when the waiter promises to eventually issue another wakeup e.g. when it releases the bit on the page. So here, wake-one semantics are implemented for those cases, using the WQ_FLAG_EXCLUSIVE flag in the waitqueue and the surrounding API's e.g. prepare_to_wait_exclusive(). I took the small liberty of adding list_for_each_entry_reverse_safe() to list.h as it generally makes sense, and gives the opportunity for fair FIFO wakeups wrapped up in a neat API. -- wli Index: wake-2.6.6-rc3-mm1/fs/buffer.c =================================================================== --- wake-2.6.6-rc3-mm1.orig/fs/buffer.c 2004-05-04 13:16:16.000000000 -0700 +++ wake-2.6.6-rc3-mm1/fs/buffer.c 2004-05-04 15:19:49.000000000 -0700 @@ -78,6 +78,30 @@ } EXPORT_SYMBOL(wake_up_buffer); +static void sync_buffer(struct buffer_head *bh) +{ + struct block_device *bd; + smp_mb(); + bd = bh->b_bdev; + if (bd) + blk_run_address_space(bd->bd_inode->i_mapping); +} + +void fastcall __lock_buffer(struct buffer_head *bh) +{ + wait_queue_head_t *wqh = bh_waitq_head(bh); + DEFINE_FILTERED_WAIT(wait, bh); + do { + prepare_to_wait_exclusive(wqh, &wait.wait, TASK_UNINTERRUPTIBLE); + if (buffer_locked(bh)) { + sync_buffer(bh); + io_schedule(); + } + } while (test_set_buffer_locked(bh)); + finish_wait(wqh, &wait.wait); +} +EXPORT_SYMBOL(__lock_buffer); + void fastcall unlock_buffer(struct buffer_head *bh) { clear_buffer_locked(bh); @@ -98,11 +122,7 @@ do { prepare_to_wait(wqh, &wait.wait, TASK_UNINTERRUPTIBLE); if (buffer_locked(bh)) { - struct block_device *bd; - smp_mb(); - bd = bh->b_bdev; - if (bd) - blk_run_address_space(bd->bd_inode->i_mapping); + sync_buffer(bh); io_schedule(); } } while (buffer_locked(bh)); Index: wake-2.6.6-rc3-mm1/include/linux/list.h =================================================================== --- wake-2.6.6-rc3-mm1.orig/include/linux/list.h 2004-04-30 15:06:48.000000000 -0700 +++ wake-2.6.6-rc3-mm1/include/linux/list.h 2004-05-04 13:16:53.000000000 -0700 @@ -413,6 +413,19 @@ pos = n, n = list_entry(n->member.next, typeof(*n), member)) /** + * list_for_each_entry_reverse_safe - iterate over list of given type safe against removal of list entry backward + * @pos: the type * to use as a loop counter. + * @n: another type * to use as temporary storage + * @head: the head for your list. + * @member: the name of the list_struct within the struct. + */ +#define list_for_each_entry_reverse_safe(pos, n, head, member) \ + for (pos = list_entry((head)->prev, typeof(*pos), member), \ + n = list_entry(pos->member.prev, typeof(*pos), member); \ + &pos->member != (head); \ + pos = n, n = list_entry(n->member.prev, typeof(*n), member)) + +/** * list_for_each_rcu - iterate over an rcu-protected list * @pos: the &struct list_head to use as a loop counter. * @head: the head for your list. Index: wake-2.6.6-rc3-mm1/include/linux/buffer_head.h =================================================================== --- wake-2.6.6-rc3-mm1.orig/include/linux/buffer_head.h 2004-04-30 15:05:52.000000000 -0700 +++ wake-2.6.6-rc3-mm1/include/linux/buffer_head.h 2004-05-04 15:13:37.000000000 -0700 @@ -170,6 +170,7 @@ struct buffer_head *alloc_buffer_head(int gfp_flags); void free_buffer_head(struct buffer_head * bh); void FASTCALL(unlock_buffer(struct buffer_head *bh)); +void FASTCALL(__lock_buffer(struct buffer_head *bh)); void ll_rw_block(int, int, struct buffer_head * bh[]); void sync_dirty_buffer(struct buffer_head *bh); void submit_bh(int, struct buffer_head *); @@ -279,8 +280,8 @@ static inline void lock_buffer(struct buffer_head *bh) { - while (test_set_buffer_locked(bh)) - __wait_on_buffer(bh); + if (test_set_buffer_locked(bh)) + __lock_buffer(bh); } #endif /* _LINUX_BUFFER_HEAD_H */ Index: wake-2.6.6-rc3-mm1/kernel/sched.c =================================================================== --- wake-2.6.6-rc3-mm1.orig/kernel/sched.c 2004-05-04 13:16:00.000000000 -0700 +++ wake-2.6.6-rc3-mm1/kernel/sched.c 2004-05-04 18:27:39.000000000 -0700 @@ -2524,9 +2524,14 @@ unsigned int mode = TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE; struct filtered_wait_queue *wait, *save; spin_lock_irqsave(&q->lock, flags); - list_for_each_entry_safe(wait, save, &q->task_list, wait.task_list) { - if (wait->key == key) - wait->wait.func(&wait->wait, mode, 0); + list_for_each_entry_reverse_safe(wait, save, &q->task_list, wait.task_list) { + int exclusive = wait->wait.flags & WQ_FLAG_EXCLUSIVE; + if (wait->key != key) + continue; + else if (!wait->wait.func(&wait->wait, mode, 0)) + continue; + else if (exclusive) + break; } spin_unlock_irqrestore(&q->lock, flags); } Index: wake-2.6.6-rc3-mm1/mm/filemap.c =================================================================== --- wake-2.6.6-rc3-mm1.orig/mm/filemap.c 2004-05-04 13:16:00.000000000 -0700 +++ wake-2.6.6-rc3-mm1/mm/filemap.c 2004-05-04 15:24:01.000000000 -0700 @@ -297,17 +297,23 @@ * at a cost of "thundering herd" phenomena during rare hash * collisions. */ -static wait_queue_head_t *page_waitqueue(struct page *page) +static wait_queue_head_t *page_waitqueue(struct page *page, int bit) { const struct zone *zone = page_zone(page); + unsigned long key = (unsigned long)page + bit; + return &zone->wait_table[hash_long(key, zone->wait_table_bits)]; +} - return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)]; +#define PAGE_KEY_SHIFT (BITS_PER_LONG - (BITS_PER_LONG == 32 ? 5 : 6)) +static void *page_key(struct page *page, unsigned long bit) +{ + return (void *)(page_to_pfn(page) | bit << PAGE_KEY_SHIFT); } void fastcall wait_on_page_bit(struct page *page, int bit_nr) { - wait_queue_head_t *waitqueue = page_waitqueue(page); - DEFINE_FILTERED_WAIT(wait, page); + wait_queue_head_t *waitqueue = page_waitqueue(page, bit_nr); + DEFINE_FILTERED_WAIT(wait, page_key(page, bit_nr)); do { prepare_to_wait(waitqueue, &wait.wait, TASK_UNINTERRUPTIBLE); @@ -338,13 +344,13 @@ */ void fastcall unlock_page(struct page *page) { - wait_queue_head_t *waitqueue = page_waitqueue(page); + wait_queue_head_t *waitqueue = page_waitqueue(page, PG_locked); smp_mb__before_clear_bit(); if (!TestClearPageLocked(page)) BUG(); smp_mb__after_clear_bit(); if (waitqueue_active(waitqueue)) - wake_up_filtered(waitqueue, page); + wake_up_filtered(waitqueue, page_key(page, PG_locked)); } EXPORT_SYMBOL(unlock_page); @@ -355,7 +361,7 @@ */ void end_page_writeback(struct page *page) { - wait_queue_head_t *waitqueue = page_waitqueue(page); + wait_queue_head_t *waitqueue = page_waitqueue(page, PG_writeback); if (!TestClearPageReclaim(page) || rotate_reclaimable_page(page)) { if (!test_clear_page_writeback(page)) @@ -363,7 +369,7 @@ smp_mb__after_clear_bit(); } if (waitqueue_active(waitqueue)) - wake_up_filtered(waitqueue, page); + wake_up_filtered(waitqueue, page_key(page, PG_writeback)); } EXPORT_SYMBOL(end_page_writeback); @@ -378,11 +384,11 @@ */ void fastcall __lock_page(struct page *page) { - wait_queue_head_t *wqh = page_waitqueue(page); - DEFINE_FILTERED_WAIT(wait, page); + wait_queue_head_t *wqh = page_waitqueue(page, PG_locked); + DEFINE_FILTERED_WAIT(wait, page_key(page, PG_locked)); while (TestSetPageLocked(page)) { - prepare_to_wait(wqh, &wait.wait, TASK_UNINTERRUPTIBLE); + prepare_to_wait_exclusive(wqh, &wait.wait, TASK_UNINTERRUPTIBLE); if (PageLocked(page)) { sync_page(page); io_schedule(); ^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [3/3] wake-one PG_locked/BH_Lock semantics 2004-05-05 6:16 ` [3/3] wake-one PG_locked/BH_Lock semantics William Lee Irwin III @ 2004-05-05 6:42 ` Michael J. Cohen 2004-05-05 9:29 ` Michael J. Cohen 0 siblings, 1 reply; 6+ messages in thread From: Michael J. Cohen @ 2004-05-05 6:42 UTC (permalink / raw) To: William Lee Irwin III; +Cc: akpm, linux-kernel tiobench numbers will be coming shortly from my test box but it appears that this one at least feels slightly better than mainline. ------ Michael On Wed, 2004-05-05 at 02:16, William Lee Irwin III wrote: > On Tue, May 04, 2004 at 11:11:21PM -0700, William Lee Irwin III wrote: > > Now, make bh's use the new wakeup primitive also. This has the bugfix > > vs. the prior version that autoremoved waitqueue wakeup functions are > > made to match autoremove API usage in __wait_event_filtered(). > > This is still grossly inefficient in that it's only necessary to wake > one waiter when the waiter promises to eventually issue another wakeup > e.g. when it releases the bit on the page. So here, wake-one semantics > are implemented for those cases, using the WQ_FLAG_EXCLUSIVE flag in > the waitqueue and the surrounding API's e.g. prepare_to_wait_exclusive(). > > I took the small liberty of adding list_for_each_entry_reverse_safe() > to list.h as it generally makes sense, and gives the opportunity for > fair FIFO wakeups wrapped up in a neat API. > > > -- wli > > Index: wake-2.6.6-rc3-mm1/fs/buffer.c > =================================================================== > --- wake-2.6.6-rc3-mm1.orig/fs/buffer.c 2004-05-04 13:16:16.000000000 -0700 > +++ wake-2.6.6-rc3-mm1/fs/buffer.c 2004-05-04 15:19:49.000000000 -0700 > @@ -78,6 +78,30 @@ > } > EXPORT_SYMBOL(wake_up_buffer); > > +static void sync_buffer(struct buffer_head *bh) > +{ > + struct block_device *bd; > + smp_mb(); > + bd = bh->b_bdev; > + if (bd) > + blk_run_address_space(bd->bd_inode->i_mapping); > +} > + > +void fastcall __lock_buffer(struct buffer_head *bh) > +{ > + wait_queue_head_t *wqh = bh_waitq_head(bh); > + DEFINE_FILTERED_WAIT(wait, bh); > + do { > + prepare_to_wait_exclusive(wqh, &wait.wait, TASK_UNINTERRUPTIBLE); > + if (buffer_locked(bh)) { > + sync_buffer(bh); > + io_schedule(); > + } > + } while (test_set_buffer_locked(bh)); > + finish_wait(wqh, &wait.wait); > +} > +EXPORT_SYMBOL(__lock_buffer); > + > void fastcall unlock_buffer(struct buffer_head *bh) > { > clear_buffer_locked(bh); > @@ -98,11 +122,7 @@ > do { > prepare_to_wait(wqh, &wait.wait, TASK_UNINTERRUPTIBLE); > if (buffer_locked(bh)) { > - struct block_device *bd; > - smp_mb(); > - bd = bh->b_bdev; > - if (bd) > - blk_run_address_space(bd->bd_inode->i_mapping); > + sync_buffer(bh); > io_schedule(); > } > } while (buffer_locked(bh)); > Index: wake-2.6.6-rc3-mm1/include/linux/list.h > =================================================================== > --- wake-2.6.6-rc3-mm1.orig/include/linux/list.h 2004-04-30 15:06:48.000000000 -0700 > +++ wake-2.6.6-rc3-mm1/include/linux/list.h 2004-05-04 13:16:53.000000000 -0700 > @@ -413,6 +413,19 @@ > pos = n, n = list_entry(n->member.next, typeof(*n), member)) > > /** > + * list_for_each_entry_reverse_safe - iterate over list of given type safe against removal of list entry backward > + * @pos: the type * to use as a loop counter. > + * @n: another type * to use as temporary storage > + * @head: the head for your list. > + * @member: the name of the list_struct within the struct. > + */ > +#define list_for_each_entry_reverse_safe(pos, n, head, member) \ > + for (pos = list_entry((head)->prev, typeof(*pos), member), \ > + n = list_entry(pos->member.prev, typeof(*pos), member); \ > + &pos->member != (head); \ > + pos = n, n = list_entry(n->member.prev, typeof(*n), member)) > + > +/** > * list_for_each_rcu - iterate over an rcu-protected list > * @pos: the &struct list_head to use as a loop counter. > * @head: the head for your list. > Index: wake-2.6.6-rc3-mm1/include/linux/buffer_head.h > =================================================================== > --- wake-2.6.6-rc3-mm1.orig/include/linux/buffer_head.h 2004-04-30 15:05:52.000000000 -0700 > +++ wake-2.6.6-rc3-mm1/include/linux/buffer_head.h 2004-05-04 15:13:37.000000000 -0700 > @@ -170,6 +170,7 @@ > struct buffer_head *alloc_buffer_head(int gfp_flags); > void free_buffer_head(struct buffer_head * bh); > void FASTCALL(unlock_buffer(struct buffer_head *bh)); > +void FASTCALL(__lock_buffer(struct buffer_head *bh)); > void ll_rw_block(int, int, struct buffer_head * bh[]); > void sync_dirty_buffer(struct buffer_head *bh); > void submit_bh(int, struct buffer_head *); > @@ -279,8 +280,8 @@ > > static inline void lock_buffer(struct buffer_head *bh) > { > - while (test_set_buffer_locked(bh)) > - __wait_on_buffer(bh); > + if (test_set_buffer_locked(bh)) > + __lock_buffer(bh); > } > > #endif /* _LINUX_BUFFER_HEAD_H */ > Index: wake-2.6.6-rc3-mm1/kernel/sched.c > =================================================================== > --- wake-2.6.6-rc3-mm1.orig/kernel/sched.c 2004-05-04 13:16:00.000000000 -0700 > +++ wake-2.6.6-rc3-mm1/kernel/sched.c 2004-05-04 18:27:39.000000000 -0700 > @@ -2524,9 +2524,14 @@ > unsigned int mode = TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE; > struct filtered_wait_queue *wait, *save; > spin_lock_irqsave(&q->lock, flags); > - list_for_each_entry_safe(wait, save, &q->task_list, wait.task_list) { > - if (wait->key == key) > - wait->wait.func(&wait->wait, mode, 0); > + list_for_each_entry_reverse_safe(wait, save, &q->task_list, wait.task_list) { > + int exclusive = wait->wait.flags & WQ_FLAG_EXCLUSIVE; > + if (wait->key != key) > + continue; > + else if (!wait->wait.func(&wait->wait, mode, 0)) > + continue; > + else if (exclusive) > + break; > } > spin_unlock_irqrestore(&q->lock, flags); > } > Index: wake-2.6.6-rc3-mm1/mm/filemap.c > =================================================================== > --- wake-2.6.6-rc3-mm1.orig/mm/filemap.c 2004-05-04 13:16:00.000000000 -0700 > +++ wake-2.6.6-rc3-mm1/mm/filemap.c 2004-05-04 15:24:01.000000000 -0700 > @@ -297,17 +297,23 @@ > * at a cost of "thundering herd" phenomena during rare hash > * collisions. > */ > -static wait_queue_head_t *page_waitqueue(struct page *page) > +static wait_queue_head_t *page_waitqueue(struct page *page, int bit) > { > const struct zone *zone = page_zone(page); > + unsigned long key = (unsigned long)page + bit; > + return &zone->wait_table[hash_long(key, zone->wait_table_bits)]; > +} > > - return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)]; > +#define PAGE_KEY_SHIFT (BITS_PER_LONG - (BITS_PER_LONG == 32 ? 5 : 6)) > +static void *page_key(struct page *page, unsigned long bit) > +{ > + return (void *)(page_to_pfn(page) | bit << PAGE_KEY_SHIFT); > } > > void fastcall wait_on_page_bit(struct page *page, int bit_nr) > { > - wait_queue_head_t *waitqueue = page_waitqueue(page); > - DEFINE_FILTERED_WAIT(wait, page); > + wait_queue_head_t *waitqueue = page_waitqueue(page, bit_nr); > + DEFINE_FILTERED_WAIT(wait, page_key(page, bit_nr)); > > do { > prepare_to_wait(waitqueue, &wait.wait, TASK_UNINTERRUPTIBLE); > @@ -338,13 +344,13 @@ > */ > void fastcall unlock_page(struct page *page) > { > - wait_queue_head_t *waitqueue = page_waitqueue(page); > + wait_queue_head_t *waitqueue = page_waitqueue(page, PG_locked); > smp_mb__before_clear_bit(); > if (!TestClearPageLocked(page)) > BUG(); > smp_mb__after_clear_bit(); > if (waitqueue_active(waitqueue)) > - wake_up_filtered(waitqueue, page); > + wake_up_filtered(waitqueue, page_key(page, PG_locked)); > } > > EXPORT_SYMBOL(unlock_page); > @@ -355,7 +361,7 @@ > */ > void end_page_writeback(struct page *page) > { > - wait_queue_head_t *waitqueue = page_waitqueue(page); > + wait_queue_head_t *waitqueue = page_waitqueue(page, PG_writeback); > > if (!TestClearPageReclaim(page) || rotate_reclaimable_page(page)) { > if (!test_clear_page_writeback(page)) > @@ -363,7 +369,7 @@ > smp_mb__after_clear_bit(); > } > if (waitqueue_active(waitqueue)) > - wake_up_filtered(waitqueue, page); > + wake_up_filtered(waitqueue, page_key(page, PG_writeback)); > } > > EXPORT_SYMBOL(end_page_writeback); > @@ -378,11 +384,11 @@ > */ > void fastcall __lock_page(struct page *page) > { > - wait_queue_head_t *wqh = page_waitqueue(page); > - DEFINE_FILTERED_WAIT(wait, page); > + wait_queue_head_t *wqh = page_waitqueue(page, PG_locked); > + DEFINE_FILTERED_WAIT(wait, page_key(page, PG_locked)); > > while (TestSetPageLocked(page)) { > - prepare_to_wait(wqh, &wait.wait, TASK_UNINTERRUPTIBLE); > + prepare_to_wait_exclusive(wqh, &wait.wait, TASK_UNINTERRUPTIBLE); > if (PageLocked(page)) { > sync_page(page); > io_schedule(); > - > 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/ ^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [3/3] wake-one PG_locked/BH_Lock semantics 2004-05-05 6:42 ` Michael J. Cohen @ 2004-05-05 9:29 ` Michael J. Cohen 0 siblings, 0 replies; 6+ messages in thread From: Michael J. Cohen @ 2004-05-05 9:29 UTC (permalink / raw) To: linux-kernel [-- Attachment #1: Type: text/plain, Size: 120 bytes --] here's the tiobenches, run several times. there was no significant deviation for -mm1 so it's not considered a problem. [-- Attachment #2: tiobench-p4m-2.6.6-rc3-mm1 --] [-- Type: text/plain, Size: 3422 bytes --] No size specified, using 1022 MB Unit information ================ File size = megabytes Blk Size = bytes Rate = megabytes per second CPU% = percentage of CPU used during the test Latency = milliseconds Lat% = percent of requests that took longer than X seconds CPU Eff = Rate divided by CPU% - throughput per cpu load Sequential Reads File Blk Num Avg Maximum Lat% Lat% CPU Identifier Size Size Thr Rate (CPU%) Latency Latency >2s >10s Eff ---------------------------- ------ ----- --- ------ ------ --------- ----------- -------- -------- ----- 2.6.6-rc3-mm1 1022 4096 1 27.59 8.909% 0.138 313.59 0.00000 0.00000 310 2.6.6-rc3-mm1 1022 4096 2 26.30 8.676% 0.293 392.22 0.00000 0.00000 303 2.6.6-rc3-mm1 1022 4096 4 24.55 8.094% 0.625 525.17 0.00000 0.00000 303 2.6.6-rc3-mm1 1022 4096 8 24.34 7.966% 1.224 1035.81 0.00000 0.00000 306 Random Reads File Blk Num Avg Maximum Lat% Lat% CPU Identifier Size Size Thr Rate (CPU%) Latency Latency >2s >10s Eff ---------------------------- ------ ----- --- ------ ------ --------- ----------- -------- -------- ----- 2.6.6-rc3-mm1 1022 4096 1 0.52 0.429% 7.444 51.22 0.00000 0.00000 122 2.6.6-rc3-mm1 1022 4096 2 0.54 0.255% 14.238 212.31 0.00000 0.00000 211 2.6.6-rc3-mm1 1022 4096 4 0.59 0.332% 25.666 234.95 0.00000 0.00000 178 2.6.6-rc3-mm1 1022 4096 8 0.60 0.334% 45.477 413.88 0.00000 0.00000 180 Sequential Writes File Blk Num Avg Maximum Lat% Lat% CPU Identifier Size Size Thr Rate (CPU%) Latency Latency >2s >10s Eff ---------------------------- ------ ----- --- ------ ------ --------- ----------- -------- -------- ----- 2.6.6-rc3-mm1 1022 4096 1 24.51 18.11% 0.146 8561.11 0.00038 0.00000 135 2.6.6-rc3-mm1 1022 4096 2 26.12 19.45% 0.253 5324.12 0.00038 0.00000 134 2.6.6-rc3-mm1 1022 4096 4 20.58 15.03% 0.613 10596.65 0.00153 0.00077 137 2.6.6-rc3-mm1 1022 4096 8 20.87 15.14% 1.143 10301.78 0.02345 0.00077 138 Random Writes File Blk Num Avg Maximum Lat% Lat% CPU Identifier Size Size Thr Rate (CPU%) Latency Latency >2s >10s Eff ---------------------------- ------ ----- --- ------ ------ --------- ----------- -------- -------- ----- 2.6.6-rc3-mm1 1022 4096 1 0.90 0.473% 0.013 0.37 0.00000 0.00000 191 2.6.6-rc3-mm1 1022 4096 2 0.85 0.418% 0.093 175.91 0.00000 0.00000 203 2.6.6-rc3-mm1 1022 4096 4 0.84 0.444% 0.020 25.46 0.00000 0.00000 188 2.6.6-rc3-mm1 1022 4096 8 0.81 0.444% 0.013 0.37 0.00000 0.00000 182 [-- Attachment #3: tiobench-p4m-2.6.6-rc3-mm1-waitone --] [-- Type: text/plain, Size: 3422 bytes --] No size specified, using 1022 MB Unit information ================ File size = megabytes Blk Size = bytes Rate = megabytes per second CPU% = percentage of CPU used during the test Latency = milliseconds Lat% = percent of requests that took longer than X seconds CPU Eff = Rate divided by CPU% - throughput per cpu load Sequential Reads File Blk Num Avg Maximum Lat% Lat% CPU Identifier Size Size Thr Rate (CPU%) Latency Latency >2s >10s Eff ---------------------------- ------ ----- --- ------ ------ --------- ----------- -------- -------- ----- 2.6.6-rc3-mm1 1022 4096 1 25.94 11.71% 0.146 370.36 0.00000 0.00000 221 2.6.6-rc3-mm1 1022 4096 2 24.96 11.34% 0.304 507.07 0.00000 0.00000 220 2.6.6-rc3-mm1 1022 4096 4 24.35 10.74% 0.626 776.48 0.00000 0.00000 227 2.6.6-rc3-mm1 1022 4096 8 24.50 10.80% 1.225 1037.49 0.00000 0.00000 227 Random Reads File Blk Num Avg Maximum Lat% Lat% CPU Identifier Size Size Thr Rate (CPU%) Latency Latency >2s >10s Eff ---------------------------- ------ ----- --- ------ ------ --------- ----------- -------- -------- ----- 2.6.6-rc3-mm1 1022 4096 1 0.58 0.579% 6.761 108.38 0.00000 0.00000 100 2.6.6-rc3-mm1 1022 4096 2 0.66 0.446% 11.394 134.13 0.00000 0.00000 147 2.6.6-rc3-mm1 1022 4096 4 0.64 0.414% 23.910 267.87 0.00000 0.00000 155 2.6.6-rc3-mm1 1022 4096 8 0.74 0.500% 39.739 379.63 0.00000 0.00000 147 Sequential Writes File Blk Num Avg Maximum Lat% Lat% CPU Identifier Size Size Thr Rate (CPU%) Latency Latency >2s >10s Eff ---------------------------- ------ ----- --- ------ ------ --------- ----------- -------- -------- ----- 2.6.6-rc3-mm1 1022 4096 1 25.92 28.26% 0.119 789.56 0.00000 0.00000 92 2.6.6-rc3-mm1 1022 4096 2 25.08 24.29% 0.236 7824.13 0.00076 0.00000 103 2.6.6-rc3-mm1 1022 4096 4 23.12 24.59% 0.532 8510.70 0.00038 0.00000 94 2.6.6-rc3-mm1 1022 4096 8 22.88 21.04% 1.018 7272.02 0.01922 0.00000 109 Random Writes File Blk Num Avg Maximum Lat% Lat% CPU Identifier Size Size Thr Rate (CPU%) Latency Latency >2s >10s Eff ---------------------------- ------ ----- --- ------ ------ --------- ----------- -------- -------- ----- 2.6.6-rc3-mm1 1022 4096 1 0.87 0.620% 0.023 19.71 0.00000 0.00000 140 2.6.6-rc3-mm1 1022 4096 2 0.87 0.632% 0.048 118.52 0.00000 0.00000 137 2.6.6-rc3-mm1 1022 4096 4 0.84 0.557% 0.020 11.00 0.00000 0.00000 150 2.6.6-rc3-mm1 1022 4096 8 0.81 0.626% 0.018 0.45 0.00000 0.00000 129 [-- Attachment #4: tiobench-p4m-2.6.6-rc3-mm1-waitone-2 --] [-- Type: text/plain, Size: 3422 bytes --] No size specified, using 1022 MB Unit information ================ File size = megabytes Blk Size = bytes Rate = megabytes per second CPU% = percentage of CPU used during the test Latency = milliseconds Lat% = percent of requests that took longer than X seconds CPU Eff = Rate divided by CPU% - throughput per cpu load Sequential Reads File Blk Num Avg Maximum Lat% Lat% CPU Identifier Size Size Thr Rate (CPU%) Latency Latency >2s >10s Eff ---------------------------- ------ ----- --- ------ ------ --------- ----------- -------- -------- ----- 2.6.6-rc3-mm1 1022 4096 1 27.70 8.979% 0.138 337.13 0.00000 0.00000 309 2.6.6-rc3-mm1 1022 4096 2 26.03 8.543% 0.295 271.51 0.00000 0.00000 305 2.6.6-rc3-mm1 1022 4096 4 23.63 7.704% 0.643 533.84 0.00000 0.00000 307 2.6.6-rc3-mm1 1022 4096 8 22.98 7.555% 1.295 1157.90 0.00000 0.00000 304 Random Reads File Blk Num Avg Maximum Lat% Lat% CPU Identifier Size Size Thr Rate (CPU%) Latency Latency >2s >10s Eff ---------------------------- ------ ----- --- ------ ------ --------- ----------- -------- -------- ----- 2.6.6-rc3-mm1 1022 4096 1 0.48 0.359% 8.204 58.92 0.00000 0.00000 132 2.6.6-rc3-mm1 1022 4096 2 0.51 0.259% 14.726 208.67 0.00000 0.00000 198 2.6.6-rc3-mm1 1022 4096 4 0.52 0.303% 29.583 256.53 0.00000 0.00000 172 2.6.6-rc3-mm1 1022 4096 8 0.59 0.283% 49.995 347.57 0.00000 0.00000 208 Sequential Writes File Blk Num Avg Maximum Lat% Lat% CPU Identifier Size Size Thr Rate (CPU%) Latency Latency >2s >10s Eff ---------------------------- ------ ----- --- ------ ------ --------- ----------- -------- -------- ----- 2.6.6-rc3-mm1 1022 4096 1 23.53 36.09% 0.135 8438.51 0.00038 0.00000 65 2.6.6-rc3-mm1 1022 4096 2 26.13 19.06% 0.235 1934.21 0.00000 0.00000 137 2.6.6-rc3-mm1 1022 4096 4 23.62 17.09% 0.539 9043.92 0.00115 0.00000 138 2.6.6-rc3-mm1 1022 4096 8 24.08 17.34% 0.985 7376.00 0.01499 0.00000 139 Random Writes File Blk Num Avg Maximum Lat% Lat% CPU Identifier Size Size Thr Rate (CPU%) Latency Latency >2s >10s Eff ---------------------------- ------ ----- --- ------ ------ --------- ----------- -------- -------- ----- 2.6.6-rc3-mm1 1022 4096 1 0.90 0.465% 0.014 0.26 0.00000 0.00000 193 2.6.6-rc3-mm1 1022 4096 2 0.87 0.468% 0.014 0.32 0.00000 0.00000 186 2.6.6-rc3-mm1 1022 4096 4 0.84 0.438% 0.013 0.09 0.00000 0.00000 191 2.6.6-rc3-mm1 1022 4096 8 0.84 0.460% 0.014 1.46 0.00000 0.00000 182 [-- Attachment #5: Type: text/plain, Size: 3422 bytes --] No size specified, using 1022 MB Unit information ================ File size = megabytes Blk Size = bytes Rate = megabytes per second CPU% = percentage of CPU used during the test Latency = milliseconds Lat% = percent of requests that took longer than X seconds CPU Eff = Rate divided by CPU% - throughput per cpu load Sequential Reads File Blk Num Avg Maximum Lat% Lat% CPU Identifier Size Size Thr Rate (CPU%) Latency Latency >2s >10s Eff ---------------------------- ------ ----- --- ------ ------ --------- ----------- -------- -------- ----- 2.6.6-rc3-mm1 1022 4096 1 27.00 8.584% 0.142 346.75 0.00000 0.00000 315 2.6.6-rc3-mm1 1022 4096 2 25.95 8.042% 0.295 413.18 0.00000 0.00000 323 2.6.6-rc3-mm1 1022 4096 4 23.97 7.407% 0.639 701.70 0.00000 0.00000 324 2.6.6-rc3-mm1 1022 4096 8 24.52 7.811% 1.218 1026.37 0.00000 0.00000 314 Random Reads File Blk Num Avg Maximum Lat% Lat% CPU Identifier Size Size Thr Rate (CPU%) Latency Latency >2s >10s Eff ---------------------------- ------ ----- --- ------ ------ --------- ----------- -------- -------- ----- 2.6.6-rc3-mm1 1022 4096 1 0.64 0.374% 6.134 132.43 0.00000 0.00000 170 2.6.6-rc3-mm1 1022 4096 2 0.67 0.262% 11.186 176.23 0.00000 0.00000 256 2.6.6-rc3-mm1 1022 4096 4 0.67 0.289% 20.849 343.38 0.00000 0.00000 230 2.6.6-rc3-mm1 1022 4096 8 0.67 0.264% 45.265 396.09 0.00000 0.00000 252 Sequential Writes File Blk Num Avg Maximum Lat% Lat% CPU Identifier Size Size Thr Rate (CPU%) Latency Latency >2s >10s Eff ---------------------------- ------ ----- --- ------ ------ --------- ----------- -------- -------- ----- 2.6.6-rc3-mm1 1022 4096 1 24.72 18.37% 0.149 9206.91 0.00038 0.00000 135 2.6.6-rc3-mm1 1022 4096 2 24.82 18.16% 0.234 2451.44 0.00038 0.00000 137 2.6.6-rc3-mm1 1022 4096 4 23.66 17.31% 0.586 8381.46 0.00153 0.00000 137 2.6.6-rc3-mm1 1022 4096 8 23.45 16.87% 0.942 7985.45 0.01423 0.00000 139 Random Writes File Blk Num Avg Maximum Lat% Lat% CPU Identifier Size Size Thr Rate (CPU%) Latency Latency >2s >10s Eff ---------------------------- ------ ----- --- ------ ------ --------- ----------- -------- -------- ----- 2.6.6-rc3-mm1 1022 4096 1 0.91 0.470% 0.013 0.27 0.00000 0.00000 193 2.6.6-rc3-mm1 1022 4096 2 0.83 0.402% 0.013 0.27 0.00000 0.00000 206 2.6.6-rc3-mm1 1022 4096 4 0.85 0.417% 0.013 0.29 0.00000 0.00000 203 2.6.6-rc3-mm1 1022 4096 8 0.81 0.388% 0.013 0.27 0.00000 0.00000 208 ^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2004-05-05 9:30 UTC | newest] Thread overview: 6+ messages (download: mbox.gz follow: Atom feed -- links below jump to the message on this page -- 2004-05-05 6:06 [0/3] filtered wakeups respun William Lee Irwin III 2004-05-05 6:08 ` William Lee Irwin III 2004-05-05 6:11 ` [2/3] filtered buffer_head wakeups William Lee Irwin III 2004-05-05 6:16 ` [3/3] wake-one PG_locked/BH_Lock semantics William Lee Irwin III 2004-05-05 6:42 ` Michael J. Cohen 2004-05-05 9:29 ` Michael J. Cohen
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox