Slide 1
Slide 1 text
Interactive
Portion
Slide 2
Slide 2 text
Unlock your
phones
Slide 3
Slide 3 text
Selfie
Sticks
Slide 4
Slide 4 text
Two
Purposes
Slide 5
Slide 5 text
Help you identify people
who like fun stuff and to
have a fun time
Slide 6
Slide 6 text
Let you know who
doesn’t like to have
fun or like fun things
Slide 7
Slide 7 text
Have you
seen this?
Slide 8
Slide 8 text
No content
Slide 9
Slide 9 text
It’s me!
Slide 10
Slide 10 text
No content
Slide 11
Slide 11 text
No content
Slide 12
Slide 12 text
No content
Slide 13
Slide 13 text
No content
Slide 14
Slide 14 text
No content
Slide 15
Slide 15 text
No content
Slide 16
Slide 16 text
Methods of
Memory
Management in
MRI
Slide 17
Slide 17 text
mmmmRuby
Slide 18
Slide 18 text
Aaron Patterson
@tenderlove
PGP Fingerprint: 4CE9 1B75 A798 28E8 6B1A A8BB 9531 70BC B4FF AFC6
Slide 19
Slide 19 text
No content
Slide 20
Slide 20 text
I am from Seattle
Slide 21
Slide 21 text
Which is not Ohio
Slide 22
Slide 22 text
!
Slide 23
Slide 23 text
"Ohio"
Slide 24
Slide 24 text
h
GitHub
Slide 25
Slide 25 text
LeGit
Slide 26
Slide 26 text
GitHub Certified®
Engineer
Slide 27
Slide 27 text
Bear Metal
Slide 28
Slide 28 text
x tenderlove
Slide 29
Slide 29 text
Please call
me "Aaron"
Slide 30
Slide 30 text
("tenderlove" is fine too)
Slide 31
Slide 31 text
I love cats!
Slide 32
Slide 32 text
Cats are the
best!
Slide 33
Slide 33 text
No content
Slide 34
Slide 34 text
No content
Slide 35
Slide 35 text
No content
Slide 36
Slide 36 text
No content
Slide 37
Slide 37 text
No content
Slide 38
Slide 38 text
I have stickers of
my cats
Slide 39
Slide 39 text
I also have
GitHub stickers!
Slide 40
Slide 40 text
I was in the news
recently
Slide 41
Slide 41 text
No content
Slide 42
Slide 42 text
Keyboards
Slide 43
Slide 43 text
No content
Slide 44
Slide 44 text
No content
Slide 45
Slide 45 text
New Ruby
Features
Slide 46
Slide 46 text
Soft Typing
Slide 47
Slide 47 text
Dynamic Typing
Slide 48
Slide 48 text
Static Typing
Slide 49
Slide 49 text
GC
Slide 50
Slide 50 text
Garbage Collector
Slide 51
Slide 51 text
Memory Terms
Slide 52
Slide 52 text
Stack Heap
Slide 53
Slide 53 text
Heap
Ruby Heap
Slide 54
Slide 54 text
GC in MRI
Slide 55
Slide 55 text
Apps in
Production
Slide 56
Slide 56 text
Scaling Issues
Slide 57
Slide 57 text
Tuning Issues
Slide 58
Slide 58 text
What I want you
to learn
Slide 59
Slide 59 text
If you don’t know
much about GC
Slide 60
Slide 60 text
If you know about
GC terminology
Slide 61
Slide 61 text
If you already know
the algorithms
Slide 62
Slide 62 text
If you already
know the new stuff
Slide 63
Slide 63 text
GC Algorithms
(in MRI)
Slide 64
Slide 64 text
Two sides of a GC
Slide 65
Slide 65 text
Collection
Algorithm
Slide 66
Slide 66 text
Allocation
Algorithm
Slide 67
Slide 67 text
Introspection API
Slide 68
Slide 68 text
Tuning Variables
Slide 69
Slide 69 text
Collection
Algorithm
Slide 70
Slide 70 text
No content
Slide 71
Slide 71 text
No content
Slide 72
Slide 72 text
What type is the
MRI collector?
Slide 73
Slide 73 text
Generational
Incremental
Mark & Sweep
Slide 74
Slide 74 text
High level:
What is a GC?
Slide 75
Slide 75 text
Tree of Objects
Root
A
B
C D
a = [
{ c: 'd' }
]
Ruby
Array
Hash
String
Symbol
Slide 76
Slide 76 text
Tree of Objects
Root
A
B
C D
a = [
{ c: 'd' }
]
a = nil
Ruby
Array
Hash
String
Symbol
Slide 77
Slide 77 text
Important Words!
Slide 78
Slide 78 text
Root Set
Slide 79
Slide 79 text
Garbage
Slide 80
Slide 80 text
Live Data
Slide 81
Slide 81 text
GC’s job:
Find unlinked nodes,
then free them
Slide 82
Slide 82 text
How to find
unlinked nodes
Slide 83
Slide 83 text
Mark & Sweep
Slide 84
Slide 84 text
2 distinct phases
Slide 85
Slide 85 text
Mark Phase
Root
D
A
B
C
F
E
H
G
Slide 86
Slide 86 text
Mark Phase
Root
D
A
B
C
F
E
H
G
Slide 87
Slide 87 text
Mark Phase
Root
D
A
B
C
F
E
H
G
Slide 88
Slide 88 text
Mark Phase
Root
D
A
B
C
F
E
H
G
Slide 89
Slide 89 text
Sweep Phase
Root
D
A
B
C
F
E
H
G
Slide 90
Slide 90 text
Sweep Phase
Root
D
A
B
C
F
Slide 91
Slide 91 text
Mark & Sweep
Very Easy!
Too Slow
Slide 92
Slide 92 text
"Stop the world"
Slide 93
Slide 93 text
Visits every
object, every time
Slide 94
Slide 94 text
Walk every object
every time
Slide 95
Slide 95 text
Generational
Slide 96
Slide 96 text
Objects die young
Slide 97
Slide 97 text
Divide objects in
to "old" and "new"
Slide 98
Slide 98 text
Generational
Root
A
B
D
C
Gen 0 Gen 1
Slide 99
Slide 99 text
Generational
Root
A
B
D
C
Gen 0 Gen 1
Slide 100
Slide 100 text
Generational
Root
A
B
D
C
Gen 0 Gen 1
Slide 101
Slide 101 text
Generational
Root
B
D
Gen 0 Gen 1
Slide 102
Slide 102 text
Generational
Root
B
D
Gen 0 Gen 1
E
F
G
Slide 103
Slide 103 text
Generational
Root
B
D
Gen 0 Gen 1
E
F
G
Slide 104
Slide 104 text
Generational
Root
B
D
Gen 0 Gen 1
E
F
G
Slide 105
Slide 105 text
Generational
Root
B
D
Gen 0 Gen 1
F
G
Slide 106
Slide 106 text
Generational
Root
B
D
Gen 0 Gen 1
F
G
Slide 107
Slide 107 text
We didn’t have to
touch "B"
Slide 108
Slide 108 text
One Slight
Problem
Slide 109
Slide 109 text
Generational
Root
B
D
Gen 0 Gen 1
Slide 110
Slide 110 text
Generational
Root
B
D
Gen 0 Gen 1
E
U
nused!
Slide 111
Slide 111 text
Remembered Set
Root
B
D
Gen 0 Gen 1
E
Slide 112
Slide 112 text
Remembered Set
Root
B
D
Gen 0 Gen 1
E
Remember!
Slide 113
Slide 113 text
Remembered Set
Root
B
D
Gen 0 Gen 1
E
Slide 114
Slide 114 text
Remembered Set
Root
B
D
Gen 0 Gen 1
E
Slide 115
Slide 115 text
Remembered Set
Root
B
D
Gen 0 Gen 1
E
Slide 116
Slide 116 text
Important Words
Slide 117
Slide 117 text
Write Barrier
Slide 118
Slide 118 text
Remembered Set
Slide 119
Slide 119 text
Generational
Fast(er)!
Not so easy
Slide 120
Slide 120 text
"Stop the world"
Slide 121
Slide 121 text
Incremental GC
Slide 122
Slide 122 text
Tri-Color
Marking
Slide 123
Slide 123 text
Object Colors
• White: will be collected
• Black: No reference to white objects, but
reachable from the root
• Gray: Reachable from root, but not yet
scanned
Slide 124
Slide 124 text
Algorithm
1. Pick an object from the gray set and move it
to the black set
2. For each object it references, move it to the
gray set
3. Repeat 1 and 2 until the gray set is empty
Slide 125
Slide 125 text
Tri-Color Marking
Root
H
G
F
C
B
D
A
E
Slide 126
Slide 126 text
Tri-Color Marking
Root
H
G
F
C
B
D
A
E
Slide 127
Slide 127 text
Tri-Color Marking
Root
H
G
F
C
B
D
A
E
Slide 128
Slide 128 text
Tri-Color Marking
Root
H
G
F
C
B
D
A
E
Slide 129
Slide 129 text
Tri-Color Marking
Root
H
G
F
C
B
D
A
E
Slide 130
Slide 130 text
No content
Slide 131
Slide 131 text
What is the
benefit?
Slide 132
Slide 132 text
We can interrupt
steps!
Slide 133
Slide 133 text
Each step can be
performed
incrementally
Slide 134
Slide 134 text
Halting time is
reduced
Slide 135
Slide 135 text
One Slight
Problem
Slide 136
Slide 136 text
Tri-Color Marking
Root
F
C
B
D
A
Slide 137
Slide 137 text
Tri-Color Marking
Root
F
C
B
D
A
G
Slide 138
Slide 138 text
Tri-Color Marking
Root
F
C
B
D
A
G
Slide 139
Slide 139 text
Tri-Color Marking
Root
F
C
B
D
A
G
Write!
Slide 140
Slide 140 text
Important Words
Slide 141
Slide 141 text
Incremental
Slide 142
Slide 142 text
Write Barrier
Slide 143
Slide 143 text
Remembered Set
Slide 144
Slide 144 text
Minimize tracing
Slide 145
Slide 145 text
Decrease halting
Slide 146
Slide 146 text
Things our GC
is not
Slide 147
Slide 147 text
Parallel
Slide 148
Slide 148 text
Real-Time
Slide 149
Slide 149 text
Compacting
Slide 150
Slide 150 text
Allocation
Algorithm
Slide 151
Slide 151 text
No content
Slide 152
Slide 152 text
No content
Slide 153
Slide 153 text
Heap layout
Slide 154
Slide 154 text
malloc isn’t free
GET IT?????
Slide 155
Slide 155 text
Large Chunk:
Page (or Slab)
Slide 156
Slide 156 text
Page memory is
contiguous
Slide 157
Slide 157 text
Each page holds a
linked list
Slide 158
Slide 158 text
Nodes are called
"slots"
Slide 159
Slide 159 text
Each slot is a
Ruby object
Slide 160
Slide 160 text
Page Layout
Page
Ruby Object
Ruby Object
Ruby Object
Ruby Object
Ruby Object
Slide 161
Slide 161 text
"Free List"
Slide 162
Slide 162 text
Find the first open
slot!
Slide 163
Slide 163 text
Move to the next
space in the free list
Slide 164
Slide 164 text
"Bump Pointer"
allocation
Slide 165
Slide 165 text
Full Pages
Object
Object
Object
Object
Page Page
Slide 166
Slide 166 text
"Eden" pages are
searched
Slide 167
Slide 167 text
GC Time
Object
Object
Object
Object
☠
Slide 168
Slide 168 text
Important Words
Slide 169
Slide 169 text
Slot
Slide 170
Slide 170 text
Page
Slide 171
Slide 171 text
Eden
Slide 172
Slide 172 text
Tomb
Slide 173
Slide 173 text
Interesting
Allocation Hacks
Slide 174
Slide 174 text
Not every object
requires allocation
Slide 175
Slide 175 text
1 Page: 16k
Slide 176
Slide 176 text
1 Object: 40 bytes
Slide 177
Slide 177 text
Pages are
"aligned"
Slide 178
Slide 178 text
Not `malloc` but
"aligned malloc"
`posix_memalign`
Slide 179
Slide 179 text
Choose "40" as a
multiple
Slide 180
Slide 180 text
Start = 40
>> start = 40
=> 40
>> (start * 1).to_s(2).rjust(10, '0')
=> "0000101000"
>> (start * 2).to_s(2).rjust(10, '0')
=> "0001010000"
>> (start * 3).to_s(2).rjust(10, '0')
=> "0001111000"
>> (start * 4).to_s(2).rjust(10, '0')
=> "0010100000"
>> (start * 5).to_s(2).rjust(10, '0')
=> "0011001000"
>> (start * 6).to_s(2).rjust(10, '0')
=> "0011110000"
>> (start * 7).to_s(2).rjust(10, '0')
=> "0100011000"
Slide 181
Slide 181 text
"0000101000"
"0001010000"
"0001111000"
"0010100000"
"0011001000"
"0011110000"
"0100011000"
Slide 182
Slide 182 text
Use these bits to
add meaning
Slide 183
Slide 183 text
Represent Integers
Without Allocation
Slide 184
Slide 184 text
Encode Number 2
>> INT_FLAG = 0x1
=> 1
>> 2.to_s(2)
=> "10"
>> ((2 << 1) | INT_FLAG).to_s(2)
=> "101"
Slide 185
Slide 185 text
Decode Number 2
>> 0b101
=> 5
>> 0b101 >> 1
=> 2
Slide 186
Slide 186 text
Biggest Fixnum
>> ((2 ** 64) - 1).to_s(2)
=>
"11111111111111111111111111111111111111111111111
11111111111111111"
>> ((2 ** 64) - 1).class
=> Bignum
>> ((2 ** 63) - 1).class
=> Bignum
>> ((2 ** 62) - 1).class
=> Fixnum
>> ((2 ** 62)).class
=> Bignum
Slide 187
Slide 187 text
Biggest Before Heap
Allocation
>> ((2 ** 64) - 1).to_s(2)
=>
"11111111111111111111111111111111111111111111111
11111111111111111"
>> ((2 ** 64) - 1).class
=> Integer
>> ((2 ** 63) - 1).class
=> Integer
>> ((2 ** 62) - 1).class
=> Integer
>> ((2 ** 62)).class
=> Integer
R
uby
2.4
Slide 188
Slide 188 text
Fixnums are
singletons
>> ((2 ** 62) - 1).object_id
=> 9223372036854775807
>> ((2 ** 62) - 1).object_id
=> 9223372036854775807
>> ((2 ** 62)).object_id
=> 70213216135940
>> ((2 ** 62)).object_id
=> 70213216117840
Slide 189
Slide 189 text
Tagged Pointer
Slide 190
Slide 190 text
Tagged Pointers
• Fixnum
• Floats
• True / False / Nil
• Symbols
Slide 191
Slide 191 text
Allocation
Problems
Slide 192
Slide 192 text
Poor Reclamation
Object
Object
Object
Object
Object
Object
Object
Object
Object
Object
Object
Object
Slide 193
Slide 193 text
Poor Reclamation
Object
Object
Object
Object
Object
Object
Object
Object
NOPE
Slide 194
Slide 194 text
Poor Reclamation
Object
Object
Object
Object
Object
Object
Object
Object
Slide 195
Slide 195 text
CoW problems
Slide 196
Slide 196 text
1 Ruby Memory Page !=
1 OS Memory Page
Slide 197
Slide 197 text
1 Ruby Page: 16k
1 OS Page: 4k
Slide 198
Slide 198 text
Object
Object
Object
Parent
Process
Child
Process
Object
Slide 199
Slide 199 text
OS Copies
1 OS Page
Slide 200
Slide 200 text
We wrote 40 bytes,
but 4kb got copied.
Slide 201
Slide 201 text
Solution:
Group Old Objects
Slide 202
Slide 202 text
Two Page Types
Probably
Old Page
Page
Object
Object
Slide 203
Slide 203 text
What is going to
be old?
Slide 204
Slide 204 text
Probably Old
class Foo
end
module Bar
CONSTANT = Object.new
def foo
"frozen string".freeze
end
end
Slide 205
Slide 205 text
Statistically
Determined
Slide 206
Slide 206 text
Efficiently use
space
Slide 207
Slide 207 text
Reduce GC time
Slide 208
Slide 208 text
CoW Friendly
Slide 209
Slide 209 text
GC Work
G
GitHub
Slide 210
Slide 210 text
No content
Slide 211
Slide 211 text
No content
Slide 212
Slide 212 text
Key
Objects
Class / Module
Empty
Slide 213
Slide 213 text
No content
Slide 214
Slide 214 text
No content
Slide 215
Slide 215 text
No content
Slide 216
Slide 216 text
No content
Slide 217
Slide 217 text
No content
Slide 218
Slide 218 text
No content
Slide 219
Slide 219 text
~17%
Smaller Heap
Slide 220
Slide 220 text
No content
Slide 221
Slide 221 text
github.com/
github/ruby
Slide 222
Slide 222 text
Future Work
Slide 223
Slide 223 text
Moving Objects
Slide 224
Slide 224 text
Poor Reclamation
Object
Object
Object
Object
Object
Object
Object
Object
NOPE
YEP
Slide 225
Slide 225 text
Stack Scanning,
Forward Pointers
Slide 226
Slide 226 text
GC
Introspection
Slide 227
Slide 227 text
Seeing GC info
Slide 228
Slide 228 text
GC.stat
{:count=>21,
:heap_allocated_pages=>87,
:heap_sorted_length=>87,
:heap_allocatable_pages=>0,
:heap_available_slots=>35460,
:heap_live_slots=>35342,
:heap_free_slots=>118,
:heap_final_slots=>0,
:heap_marked_slots=>22207,
:heap_eden_pages=>87,
:heap_tomb_pages=>0,
:total_allocated_pages=>87,
:total_freed_pages=>0,
:total_allocated_objects=>208577,
:total_freed_objects=>173235,
:malloc_increase_bytes=>5152,
:malloc_increase_bytes_limit=>16777216,
:minor_gc_count=>19,
:major_gc_count=>2,
:remembered_wb_unprotected_objects=>189,
:remembered_wb_unprotected_objects_limit=>374,
:old_objects=>21977,
:old_objects_limit=>39876,
:oldmalloc_increase_bytes=>485600,
:oldmalloc_increase_bytes_limit=>16777216}
Slide 229
Slide 229 text
GC Performance
Slide 230
Slide 230 text
GC::Profiler
>> GC::Profiler.enable
=> nil
>> GC::Profiler.report
=> nil
>> GC.start
=> nil
>> GC::Profiler.report
GC 22 invokes.
Index Invoke Time(sec) Use Size(byte) Total Size(byte) Total Object
GC Time(ms)
1 0.143 906920 1419840 35496
3.72500000000000586198
=> nil
>> GC.start
=> nil
>> GC::Profiler.report
GC 23 invokes.
Index Invoke Time(sec) Use Size(byte) Total Size(byte) Total Object
GC Time(ms)
1 0.143 906920 1419840 35496
3.72500000000000586198
2 0.148 906920 1419840 35496
3.47800000000000864020
Slide 231
Slide 231 text
GC::Profiler.enable
Slide 232
Slide 232 text
GC::Profiler.report
Slide 233
Slide 233 text
Heap Inspection
Slide 234
Slide 234 text
ObjectSpace.dump_all
Slide 235
Slide 235 text
ObjectSpace.dump
>> require 'objspace'
=> false
>> x = Object.new
=> #
>> ObjectSpace.dump x
=> "{\"address\":\"0x007fbcd09334a8\",
\"type\":\"OBJECT\", \"class\":
\"0x007fbcd08dd878\", \"ivars\":0,
\"memsize\":40, \"flags\":
{\"wb_protected\":true}}\n"
Slide 236
Slide 236 text
ObjectSpace.dump
>> x = Object.new
=> #
>> JSON.parse(ObjectSpace.dump(x))['flags']
=> {"wb_protected"=>true}
>> GC.start
=> nil
>> JSON.parse(ObjectSpace.dump(x))['flags']
=> {"wb_protected"=>true}
>> GC.start
=> nil
>> JSON.parse(ObjectSpace.dump(x))['flags']
=> {"wb_protected"=>true}
>> GC.start
=> nil
>> JSON.parse(ObjectSpace.dump(x))['flags']
=> {"wb_protected"=>true, "old"=>true, "uncollectible"=>true,
"marked"=>true}
Slide 237
Slide 237 text
Object have 3
generations
Slide 238
Slide 238 text
ObjectSpace
Slide 239
Slide 239 text
trace_object_allocations
Slide 240
Slide 240 text
GC Tuning
Slide 241
Slide 241 text
RUBY_GC_HEAP
_FREE_SLOTS
Number of free slots available after a GC
Slide 242
Slide 242 text
RUBY_GC_HEAP
_INIT_SLOTS
Number of free slots to initialize the GC with
Slide 243
Slide 243 text
RUBY_GC_HEAP_GR
OWTH_MAX_SLOTS
Never grow more than this many objects
Slide 244
Slide 244 text
THANKS!
Slide 245
Slide 245 text
No content