Skip to contentSkip to conversation

iozone regression with 2.6.29-rc6

DISCUSSION Lin Ming 4 participants 6 messages Feb 27, 2009 – Mar 2, 2009

Subsystems (2)
lore
Share & tools
lore

Lin Ming

author Acked-by Peter Zijlstra
#0 lore raw

bisect locates below commits,

commit 1cf6e7d83bf334cc5916137862c920a97aabc018

Author: Nick Piggin <[email protected]>
Date:   Wed Feb 18 14:48:18 2009 -0800

mm: task dirty accounting fix

YAMAMOTO-san noticed that task_dirty_inc doesn't seem to be called properly for cases where set_page_dirty is not used to dirty a page (eg. mark_buffer_dirty).

Additionally, there is some inconsistency about when task_dirty_inc is called. It is used for dirty balancing, however it even gets called for __set_page_dirty_no_writeback.

So rather than increment it in a set_page_dirty wrapper, move it down to exactly where the dirty page accounting stats are incremented.

below data in parenthesis is the result after above commit reverted, for example, -10% (+2%) means,
iozone has ~10% regression with 2.6.29-rc6 compared with 2.6.29-rc5. and
iozone has ~2% improvement with 2.6.29-rc6-revert-1cf6e7d compared with 2.6.29-rc5.

4P dual-core HT 2P qual-core 2P qual-core HT
tulsa stockley Nehalem
--------------------------------------------------------
iozone-rewrite -10% (+2%) -8% (0%) -10% (-7%)
iozone-rand-write -50% (0%) -20% (+10%)
iozone-read -13% (0%)
iozone-write -28% (-1%)
iozone-reread -5% (-1%)
iozone-mmap-read -7% (+2%)
iozone-mmap-reread -7% (+2%)
iozone-mmap-rand-read -7% (+3%)
iozone-mmap-rand-write -5% (0%)

Lin Ming

#1 lore raw
Quoted 32 lines · Lin Ming

bisect locates below commits,

commit 1cf6e7d83bf334cc5916137862c920a97aabc018

Author: Nick Piggin <[email protected]>
Date:   Wed Feb 18 14:48:18 2009 -0800

mm: task dirty accounting fix

YAMAMOTO-san noticed that task_dirty_inc doesn't seem to be called properly for cases where set_page_dirty is not used to dirty a page (eg. mark_buffer_dirty).

Additionally, there is some inconsistency about when task_dirty_inc is called. It is used for dirty balancing, however it even gets called for __set_page_dirty_no_writeback.

So rather than increment it in a set_page_dirty wrapper, move it down to exactly where the dirty page accounting stats are incremented.

below data in parenthesis is the result after above commit reverted, for example, -10% (+2%) means,
iozone has ~10% regression with 2.6.29-rc6 compared with 2.6.29-rc5. and
iozone has ~2% improvement with 2.6.29-rc6-revert-1cf6e7d compared with 2.6.29-rc5.

Quoted 12 lines

4P dual-core HT 2P qual-core 2P qual-core HT
tulsa stockley Nehalem
--------------------------------------------------------
iozone-rewrite -10% (+2%) -8% (0%) -10% (-7%)
iozone-rand-write -50% (0%) -20% (+10%)
iozone-read -13% (0%)
iozone-write -28% (-1%)
iozone-reread -5% (-1%)
iozone-mmap-read -7% (+2%)
iozone-mmap-reread -7% (+2%)
iozone-mmap-rand-read -7% (+3%)
iozone-mmap-rand-write -5% (0%)

Ugh, that's unexpected..

So 'better' accounting leads to worse performance, which would indicate we throttle more.

I take it you machine has gobs of memory.

Does something like the below help any?

---
Subject: mm: bdi: tweak task dirty penalty
From: Peter Zijlstra <[email protected]>
Date: Fri Feb 27 10:41:22 CET 2009

Penalizing heavy dirtiers with 1/8-th the total dirty limit might be rather excessive on large memory machines. Use sqrt to scale it sub-linearly.

Update the comment while we're there.

---
 mm/page-writeback.c |   12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)
mm/page-writeback.c +8 -4
Index: linux-2.6/mm/page-writeback.c
===================================================================
@@ -293,17 +293,21 @@ static inline void task_dirties_fraction
 }
 
 /*
- * scale the dirty limit
+ * Task specific dirty limit:
  *
- * task specific dirty limit:
+ *   dirty -= 8 * sqrt(dirty) * p_{t}
  *
- *   dirty -= (dirty/8) * p_{t}
+ * Penalize tasks that dirty a lot of pages by lowering their dirty limit. This
+ * avoids infrequent dirtiers from getting stuck in this other guys dirty
+ * pages.
+ *
+ * Use a sub-linear function to scale the penalty, we only need a little room.
  */
 static void task_dirty_limit(struct task_struct *tsk, long *pdirty)
 {
 	long numerator, denominator;
 	long dirty = *pdirty;
-	u64 inv = dirty >> 3;
+	u64 inv = 8*int_sqrt(dirty);
 
 	task_dirties_fraction(tsk, &numerator, &denominator);
 	inv *= numerator;
#2 lore raw
Quoted 46 lines · Peter Zijlstra
Quoted 32 lines · Lin Ming

bisect locates below commits,

commit 1cf6e7d83bf334cc5916137862c920a97aabc018

Author: Nick Piggin <[email protected]>
Date:   Wed Feb 18 14:48:18 2009 -0800

mm: task dirty accounting fix

YAMAMOTO-san noticed that task_dirty_inc doesn't seem to be called properly for cases where set_page_dirty is not used to dirty a page (eg. mark_buffer_dirty).

Additionally, there is some inconsistency about when task_dirty_inc is called. It is used for dirty balancing, however it even gets called for __set_page_dirty_no_writeback.

So rather than increment it in a set_page_dirty wrapper, move it down to exactly where the dirty page accounting stats are incremented.

below data in parenthesis is the result after above commit reverted, for example, -10% (+2%) means,
iozone has ~10% regression with 2.6.29-rc6 compared with 2.6.29-rc5. and
iozone has ~2% improvement with 2.6.29-rc6-revert-1cf6e7d compared with 2.6.29-rc5.

Quoted 12 lines

4P dual-core HT 2P qual-core 2P qual-core HT
tulsa stockley Nehalem
--------------------------------------------------------
iozone-rewrite -10% (+2%) -8% (0%) -10% (-7%)
iozone-rand-write -50% (0%) -20% (+10%)
iozone-read -13% (0%)
iozone-write -28% (-1%)
iozone-reread -5% (-1%)
iozone-mmap-read -7% (+2%)
iozone-mmap-reread -7% (+2%)
iozone-mmap-rand-read -7% (+3%)
iozone-mmap-rand-write -5% (0%)

Quoted 8 lines

Ugh, that's unexpected..

So 'better' accounting leads to worse performance, which would indicate we throttle more.

I take it you machine has gobs of memory.

Does something like the below help any?

Shall we revert this for 2.6.29, then? And try to improve it in the next cycle? Are we looking at a several more weeks before 2.6.29, or do we prefer not to try tweaking heuristics at this point?

Quoted 46 lines

---
Subject: mm: bdi: tweak task dirty penalty
From: Peter Zijlstra <[email protected]>
Date: Fri Feb 27 10:41:22 CET 2009

Penalizing heavy dirtiers with 1/8-th the total dirty limit might be rather excessive on large memory machines. Use sqrt to scale it sub-linearly.

Update the comment while we're there.

---
 mm/page-writeback.c |   12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)

Index: linux-2.6/mm/page-writeback.c
===================================================================

mm/page-writeback.c +8 -4
@@ -293,17 +293,21 @@ static inline void task_dirties_fraction
 }
 
 /*
- * scale the dirty limit
+ * Task specific dirty limit:
  *
- * task specific dirty limit:
+ *   dirty -= 8 * sqrt(dirty) * p_{t}
  *
- *   dirty -= (dirty/8) * p_{t}
+ * Penalize tasks that dirty a lot of pages by lowering their dirty limit. This
+ * avoids infrequent dirtiers from getting stuck in this other guys dirty
+ * pages.
+ *
+ * Use a sub-linear function to scale the penalty, we only need a little room.
  */
 static void task_dirty_limit(struct task_struct *tsk, long *pdirty)
 {
 	long numerator, denominator;
 	long dirty = *pdirty;
-	u64 inv = dirty >> 3;
+	u64 inv = 8*int_sqrt(dirty);
 
 	task_dirties_fraction(tsk, &numerator, &denominator);
 	inv *= numerator;
author
#3 lore raw
In reply to Peter Zijlstra
Quoted 46 lines · Peter Zijlstra
Quoted 32 lines · Lin Ming

bisect locates below commits,

commit 1cf6e7d83bf334cc5916137862c920a97aabc018

Author: Nick Piggin <[email protected]>
Date:   Wed Feb 18 14:48:18 2009 -0800

mm: task dirty accounting fix

YAMAMOTO-san noticed that task_dirty_inc doesn't seem to be called properly for cases where set_page_dirty is not used to dirty a page (eg. mark_buffer_dirty).

Additionally, there is some inconsistency about when task_dirty_inc is called. It is used for dirty balancing, however it even gets called for __set_page_dirty_no_writeback.

So rather than increment it in a set_page_dirty wrapper, move it down to exactly where the dirty page accounting stats are incremented.

below data in parenthesis is the result after above commit reverted, for example, -10% (+2%) means,
iozone has ~10% regression with 2.6.29-rc6 compared with 2.6.29-rc5. and
iozone has ~2% improvement with 2.6.29-rc6-revert-1cf6e7d compared with 2.6.29-rc5.

Quoted 12 lines

4P dual-core HT 2P qual-core 2P qual-core HT
tulsa stockley Nehalem
--------------------------------------------------------
iozone-rewrite -10% (+2%) -8% (0%) -10% (-7%)
iozone-rand-write -50% (0%) -20% (+10%)
iozone-read -13% (0%)
iozone-write -28% (-1%)
iozone-reread -5% (-1%)
iozone-mmap-read -7% (+2%)
iozone-mmap-reread -7% (+2%)
iozone-mmap-rand-read -7% (+3%)
iozone-mmap-rand-write -5% (0%)

Quoted 8 lines

Ugh, that's unexpected..

So 'better' accounting leads to worse performance, which would indicate we throttle more.

I take it you machine has gobs of memory.

Does something like the below help any?

It helps some as below test result,
The data in second parenthesis means 2.6.29-rc6-with-peter's-patch compared with 2.6.29-rc5.

4P dual-core HT 2P qual-core 2P qual-core HT
tulsa stockley Nehalem
--------------------------------------------------------
iozone-rewrite -10% (+2%)(-3%) -8% (0%)(0%) -10% (-7%)(-2%)
iozone-rand-write -50% (0%)(-10%) -20% (+10%)(+3%)
iozone-read -13% (0%)(-8%)
iozone-write -28% (-1%)(+35%)
iozone-reread -5% (-1%)(-1%)
iozone-mmap-read -7% (+2%)(-7%)
iozone-mmap-reread -7% (+2%)(-7%)
iozone-mmap-rand-read -7% (+3%)(-7%)
iozone-mmap-rand-write -5% (0%)(+27%)

Lin Ming

Quoted 48 lines

---
Subject: mm: bdi: tweak task dirty penalty
From: Peter Zijlstra <[email protected]>
Date: Fri Feb 27 10:41:22 CET 2009

Penalizing heavy dirtiers with 1/8-th the total dirty limit might be rather excessive on large memory machines. Use sqrt to scale it sub-linearly.

Update the comment while we're there.

---
 mm/page-writeback.c |   12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)

Index: linux-2.6/mm/page-writeback.c
===================================================================

mm/page-writeback.c +8 -4
@@ -293,17 +293,21 @@ static inline void task_dirties_fraction
 }
 
 /*
- * scale the dirty limit
+ * Task specific dirty limit:
  *
- * task specific dirty limit:
+ *   dirty -= 8 * sqrt(dirty) * p_{t}
  *
- *   dirty -= (dirty/8) * p_{t}
+ * Penalize tasks that dirty a lot of pages by lowering their dirty limit. This
+ * avoids infrequent dirtiers from getting stuck in this other guys dirty
+ * pages.
+ *
+ * Use a sub-linear function to scale the penalty, we only need a little room.
  */
 static void task_dirty_limit(struct task_struct *tsk, long *pdirty)
 {
 	long numerator, denominator;
 	long dirty = *pdirty;
-	u64 inv = dirty >> 3;
+	u64 inv = 8*int_sqrt(dirty);
 
 	task_dirties_fraction(tsk, &numerator, &denominator);
 	inv *= numerator;
#4 lore raw
Quoted 60 lines · Ming
Quoted 46 lines · Peter Zijlstra
Quoted 32 lines · Lin Ming

bisect locates below commits,

commit 1cf6e7d83bf334cc5916137862c920a97aabc018

Author: Nick Piggin <[email protected]>
Date:   Wed Feb 18 14:48:18 2009 -0800

mm: task dirty accounting fix

YAMAMOTO-san noticed that task_dirty_inc doesn't seem to be called properly for cases where set_page_dirty is not used to dirty a page (eg. mark_buffer_dirty).

Additionally, there is some inconsistency about when task_dirty_inc is called. It is used for dirty balancing, however it even gets called for __set_page_dirty_no_writeback.

So rather than increment it in a set_page_dirty wrapper, move it down to exactly where the dirty page accounting stats are incremented.

below data in parenthesis is the result after above commit reverted, for example, -10% (+2%) means,
iozone has ~10% regression with 2.6.29-rc6 compared with 2.6.29-rc5. and
iozone has ~2% improvement with 2.6.29-rc6-revert-1cf6e7d compared with 2.6.29-rc5.

Quoted 12 lines

4P dual-core HT 2P qual-core 2P qual-core HT
tulsa stockley Nehalem
--------------------------------------------------------
iozone-rewrite -10% (+2%) -8% (0%) -10% (-7%)
iozone-rand-write -50% (0%) -20% (+10%)
iozone-read -13% (0%)
iozone-write -28% (-1%)
iozone-reread -5% (-1%)
iozone-mmap-read -7% (+2%)
iozone-mmap-reread -7% (+2%)
iozone-mmap-rand-read -7% (+3%)
iozone-mmap-rand-write -5% (0%)

Quoted 8 lines

Ugh, that's unexpected..

So 'better' accounting leads to worse performance, which would indicate we throttle more.

I take it you machine has gobs of memory.

Does something like the below help any?

It helps some as below test result,
The data in second parenthesis means 2.6.29-rc6-with-peter's-patch compared with 2.6.29-rc5.

Quoted 12 lines

4P dual-core HT 2P qual-core 2P qual-core HT
tulsa stockley Nehalem
--------------------------------------------------------
iozone-rewrite -10% (+2%)(-3%) -8% (0%)(0%) -10% (-7%)(-2%)
iozone-rand-write -50% (0%)(-10%) -20% (+10%)(+3%)
iozone-read -13% (0%)(-8%)
iozone-write -28% (-1%)(+35%)
iozone-reread -5% (-1%)(-1%)
iozone-mmap-read -7% (+2%)(-7%)
iozone-mmap-reread -7% (+2%)(-7%)
iozone-mmap-rand-read -7% (+3%)(-7%)
iozone-mmap-rand-write -5% (0%)(+27%)

Thanks, Lin Ming. To better understand the situation, would you please provide the iozone command and memory info about the servers?

Thanks,
Fengguang

author
#5 lore raw
Quoted 74 lines · Fengguang
Quoted 60 lines · Ming
Quoted 46 lines · Peter Zijlstra
Quoted 32 lines · Lin Ming

bisect locates below commits,

commit 1cf6e7d83bf334cc5916137862c920a97aabc018

Author: Nick Piggin <[email protected]>
Date:   Wed Feb 18 14:48:18 2009 -0800

mm: task dirty accounting fix

YAMAMOTO-san noticed that task_dirty_inc doesn't seem to be called properly for cases where set_page_dirty is not used to dirty a page (eg. mark_buffer_dirty).

Additionally, there is some inconsistency about when task_dirty_inc is called. It is used for dirty balancing, however it even gets called for __set_page_dirty_no_writeback.

So rather than increment it in a set_page_dirty wrapper, move it down to exactly where the dirty page accounting stats are incremented.

below data in parenthesis is the result after above commit reverted, for example, -10% (+2%) means,
iozone has ~10% regression with 2.6.29-rc6 compared with 2.6.29-rc5. and
iozone has ~2% improvement with 2.6.29-rc6-revert-1cf6e7d compared with 2.6.29-rc5.

Quoted 12 lines

4P dual-core HT 2P qual-core 2P qual-core HT
tulsa stockley Nehalem
--------------------------------------------------------
iozone-rewrite -10% (+2%) -8% (0%) -10% (-7%)
iozone-rand-write -50% (0%) -20% (+10%)
iozone-read -13% (0%)
iozone-write -28% (-1%)
iozone-reread -5% (-1%)
iozone-mmap-read -7% (+2%)
iozone-mmap-reread -7% (+2%)
iozone-mmap-rand-read -7% (+3%)
iozone-mmap-rand-write -5% (0%)

Quoted 8 lines

Ugh, that's unexpected..

So 'better' accounting leads to worse performance, which would indicate we throttle more.

I take it you machine has gobs of memory.

Does something like the below help any?

It helps some as below test result,
The data in second parenthesis means 2.6.29-rc6-with-peter's-patch compared with 2.6.29-rc5.

Quoted 12 lines

4P dual-core HT 2P qual-core 2P qual-core HT
tulsa stockley Nehalem
--------------------------------------------------------
iozone-rewrite -10% (+2%)(-3%) -8% (0%)(0%) -10% (-7%)(-2%)
iozone-rand-write -50% (0%)(-10%) -20% (+10%)(+3%)
iozone-read -13% (0%)(-8%)
iozone-write -28% (-1%)(+35%)
iozone-reread -5% (-1%)(-1%)
iozone-mmap-read -7% (+2%)(-7%)
iozone-mmap-reread -7% (+2%)(-7%)
iozone-mmap-rand-read -7% (+3%)(-7%)
iozone-mmap-rand-write -5% (0%)(+27%)

Quoted 2 lines

Thanks, Lin Ming. To better understand the situation, would you please provide the iozone command and memory info about the servers?

iozone -i 0 -i 1 -i 2 -i 3 -i 4 -r 4k -s 64k -s 512m -s 1200m -b tmp.xls iozone -B -r 4k -s 64k -s 512m -s 1200m -b tmp.xls

4P dual-core HT tulsa /proc/meminfo
------------------------------------

MemTotal:        8189476 kB
MemFree:         7915884 kB
Buffers:           32676 kB
Cached:           139784 kB
SwapCached:            0 kB
Active:            35852 kB
Inactive:         145564 kB
Active(anon):       9048 kB
Inactive(anon):        0 kB
Active(file):      26804 kB
Inactive(file):   145564 kB
Unevictable:           0 kB
Mlocked:               0 kB
SwapTotal:             0 kB
SwapFree:              0 kB
Dirty:               140 kB
Writeback:             0 kB
AnonPages:          9068 kB
Mapped:             4084 kB
Slab:              31180 kB
SReclaimable:      14800 kB
SUnreclaim:        16380 kB
PageTables:         1236 kB
NFS_Unstable:          0 kB
Bounce:                0 kB
WritebackTmp:          0 kB
CommitLimit:     4094736 kB
Committed_AS:   18446744073709545248 kB
VmallocTotal:   34359738367 kB
VmallocUsed:      286416 kB
VmallocChunk:   34359449467 kB
HugePages_Total:       0
HugePages_Free:        0
HugePages_Rsvd:        0
HugePages_Surp:        0
Hugepagesize:       2048 kB
DirectMap4k:        4096 kB
DirectMap2M:     8380416 

2P qual-core stockley /proc/meminfo
-----------------------------------

MemTotal:        8167260 kB
MemFree:         7951068 kB
Buffers:           30988 kB
Cached:           123772 kB
SwapCached:            0 kB
Active:            34744 kB
Inactive:         129296 kB
Active(anon):       9412 kB
Inactive(anon):        0 kB
Active(file):      25332 kB
Inactive(file):   129296 kB
Unevictable:           0 kB
Mlocked:               0 kB
SwapTotal:             0 kB
SwapFree:              0 kB
Dirty:               272 kB
Writeback:             0 kB
AnonPages:          9408 kB
Mapped:             4072 kB
Slab:              21760 kB
SReclaimable:      10612 kB
SUnreclaim:        11148 kB
PageTables:         1260 kB
NFS_Unstable:          0 kB
Bounce:                0 kB
WritebackTmp:          0 kB
CommitLimit:     4083628 kB
Committed_AS:   18446744073709540996 kB
VmallocTotal:   34359738367 kB
VmallocUsed:       11216 kB
VmallocChunk:   34359726459 kB
HugePages_Total:       0
HugePages_Free:        0
HugePages_Rsvd:        0
HugePages_Surp:        0
Hugepagesize:       2048 kB
DirectMap4k:        7168 kB
DirectMap2M:     8380416 kB

2P qual-core HT Nehalem /proc/meminfo
-------------------------------------

MemTotal:        6113044 kB
MemFree:         5947476 kB
Buffers:           17056 kB
Cached:            44280 kB
SwapCached:            0 kB
Active:            31268 kB
Inactive:          40300 kB
Active(anon):      10308 kB
Inactive(anon):        0 kB
Active(file):      20960 kB
Inactive(file):    40300 kB
Unevictable:           0 kB
Mlocked:               0 kB
SwapTotal:             0 kB
SwapFree:              0 kB
Dirty:               280 kB
Writeback:             0 kB
AnonPages:         10328 kB
Mapped:             5096 kB
Slab:              27844 kB
SReclaimable:      10988 kB
SUnreclaim:        16856 kB
PageTables:         2204 kB
NFS_Unstable:          0 kB
Bounce:                0 kB
WritebackTmp:          0 kB
CommitLimit:     3056520 kB
Committed_AS:   18446744073709542732 kB
VmallocTotal:   34359738367 kB
VmallocUsed:      286428 kB
VmallocChunk:   34359451259 kB
HugePages_Total:       0
HugePages_Free:        0
HugePages_Rsvd:        0
HugePages_Surp:        0
Hugepagesize:       2048 kB
DirectMap4k:        8192 kB
DirectMap2M:     6275072 kB

Lin Ming

Quoted 3 lines

Thanks,
Fengguang