OSS iPedia

CLIENT数 (eu)

計測環境

800

1600

2400

3200

4000

4800

5600

6400

7200

8000

CPU=1

110.2

CPU=2

110.3

221.0

283.4

279.4

281.5

277.4

277.6

CPU=4

110.5

220.9

331.3

440.8

509.1

509.2

507.4

CPU=8

110.5

221.0

331.2

442.1

552.1

651.5

693.0

681.9

664.8

666.5

CPU=12

221.2

331.6

441.3

512.9

546.1

543.1

512.7

483.2

CPU=16

221.2

332.0

442.3

507.6

527.2

516.3

499.3

487.5

479.6

CPU数=8 MEMORYサイズ=16GB CLIENT=5600

CPU: P4 / Xeon, speed 2801.8 MHz (estimated)

Counted MEMORY_CANCEL events (cancelled requesets in data cache address control unit) with a unit mask of 0x08 (conflicts due to 64k aliasing) count 3000

samples

%

app name

symbol name

1

10222096

7.2162

postgres

AllocSetAlloc

2

5887663

4.1563

libc-2.3.4.so

memcpy

3

5828560

4.1146

postgres

element_alloc

4

4518085

3.1895

postgres

HeapTupleSatisfiesSnapshot

5

4054838

2.8625

postgres

SearchCatCache

6

3112045

2.1969

postgres

hash_search

7

2972878

2.0987

postgres

LWLockAcquire

8

2700978

1.9067

libc-2.3.4.so

malloc

9

2020607

1.4264

postgres

nocachegetattr

10

2001770

1.4131

libc-2.3.4.so

malloc_consolidate

CPU数=12 MEMORYサイズ=16GB CLIENT=5600

CPU: P4 / Xeon, speed 2801.8 MHz (estimated)

Counted BSQ_CACHE_REFERENCE events (cache references seen by the bus unit) with a unit mask of 0x200 (read 3rd level cache miss) count 3000

samples

%

app name

symbol name

1

1758277

15.0155

postgres

HeapTupleSatisfiesSnapshot

2

780948

6.6692

postgres

s_lock

3

656990

5.6106

postgres

LWLockAcquire

4

474467

4.0519

vmlinux

LKST_ETYPE_SYSV_IPC_SEMOP_HEADER_hook

5

349292

2.9829

postgres

hash_search

6

292779

2.5003

postgres

wchareq

7

282569

2.4131

postgres

PinBuffer

8

243217

2.077

postgres

SearchCatCache

9

230091

1.9649

postgres

slot_deform_tuple

10

226121

1.931

vmlinux

kmem_cache_create

$ head -26 */summary.*|grep -v warning|less

==> U2800-HTOFF/summary.out <==

CPU: P4 / Xeon, speed 3591.64 MHz (estimated)

Counted GLOBAL_POWER_EVENTS events (time during which processor is not stopped) with a unit mask of 0x01 (mandatory) count 100000

samples % app name symbol name

6673106 8.5252 mysqld btr_search_guess_on_hash

6325149 8.0806 mysqld pthread_mutex_trylock

6105048 7.7994 mysqld rec_get_offsets_func

5481538 7.0029 mysqld memcpy

4593693 5.8686 mysqld __pthread_unlock

2842017 3.6308 mysqld row_search_for_mysql

1841049 2.3520 mysqld cmp_dtuple_rec_with_match

1274417 1.6281 mysqld btr_cur_search_to_nth_level

1119738 1.4305 mysqld buf_page_get_known_nowait

==> U2800-HTON/summary.out <==

CPU: P4 / Xeon with 2 hyper-threads, speed 3591.94 MHz (estimated)

Counted GLOBAL_POWER_EVENTS events (time during which processor is not stopped) with a unit mask of 0x01 (mandatory) count 100000

samples % app name symbol name

14939404 7.8026 mysqld rec_get_offsets_func

13518106 7.0602 mysqld pthread_mutex_trylock

11911766 6.2213 mysqld btr_search_guess_on_hash

10490744 5.4791 mysqld memcpy

8590047 4.4864 mysqld __pthread_unlock

7218939 3.7703 mysqld row_search_for_mysql

6278792 3.2793 mysqld mutex_spin_wait

5520007 2.8830 mysqld cmp_dtuple_rec_with_match

3702504 1.9337 mysqld btr_cur_search_to_nth_level

3625952 1.8938 mysqld mtr_memo_slot_release

#define TAS(lock) tas(lock)

static __inline__ int

tas(volatile slock_t *lock)

{

register slock_t _res = 1;

/*

* Use a non-locking test before asserting the bus lock. Note that the

* extra test appears to be a small loss on some x86 platforms and a small

* win on others; it's by no means clear that we should keep it.

*/

__asm__ __volatile__(

" cmpb $0,%1 \n"

" jne 1f \n"

" lock \n"

" xchgb %0,%1 \n"

"1: \n"

: "+q"(_res), "+m"(*lock)

:

: "memory", "cc");

return (int) _res;

}

while (TAS(lock))

{

/* CPU-specific delay each time through the loop */

SPIN_DELAY();

/* Block the process every spins_per_delay tries */

if (++spins >= spins_per_delay)

{

if (++delays > NUM_DELAYS)

s_lock_stuck(lock, file, line);

if (cur_delay == 0) /* first time to delay? */

cur_delay = MIN_DELAY_MSEC;

pg_usleep(cur_delay * 1000L);

/* increase delay by a random fraction between 1X and 2X */

cur_delay += (int) (cur_delay *

(((double) random()) / ((double) MAX_RANDOM_VALUE)) + 0.5);

/* wrap back to minimum delay when max is exceeded */

if (cur_delay > MAX_DELAY_MSEC)

cur_delay = MIN_DELAY_MSEC;

spins = 0;

}

}

$ diff -up s_lock.c.orig s_lock.c

--- s_lock.c.orig 2005-11-23 03:23:19.000000000 +0900

+++ s_lock.c 2006-02-27 19:00:39.626059832 +0900

@@ -99,6 +99,12 @@ s_lock(volatile slock_t *lock, const cha

while (TAS(lock))

{

+ /* This double while-loop does not lock the memory bus. */

+ /* The CPU will access the CPU cache of *lock variable */

+ /* instead of the main memory. */

+ /* So this implementation scales well. */

+ while (*lock)

+ {

/* CPU-specific delay each time through the loop */

SPIN_DELAY();

@@ -127,6 +133,7 @@ s_lock(volatile slock_t *lock, const cha

spins = 0;

}

+ }

}

/*

while (mutex_get_lock_word(mutex) != 0 && i < SYNC_SPIN_ROUNDS)

{

if (srv_spin_wait_delay)

{

ut_delay(ut_rnd_interval(0, srv_spin_wait_delay));

}

i++;

}

$ diff -pu ut0ut.c.orig ut0ut.c

--- ut0ut.c.orig 2005-10-17 10:27:43.000000000 +0900

+++ ut0ut.c 2006-02-28 11:59:16.777840496 +0900

@@ -290,6 +290,13 @@ ut_delay(

j = 0;

for (i = 0; i < delay * 50; i++) {

+ /* When executing a spin-wait loop on the Hyper-Threading

+ processor, the processor can suffer a severe performance

+ penalty. The pause instruction provides a hint to the

+ processor. Please refer IA-32 Intel Architecture

+ Software Developers Manual, Vol 3. */

+ __asm__ __volatile__(

+ "pause; \n");

j += i;

}